Technology Acceptance of Future Decision Makers

Master’s Degree Project in Innovation and Industrial Management

Master’s Degree Project in Management

Technology Acceptance of Future Decision Makers

An Investigation of Behavioral Intention Toward Using

Artificial Intelligence Tools in Decision Making

Author:

Emelie Gunnarsson

GRADUATE SCHOOL

Master of Science in Innovation and Industrial Management

Supervisor: Ph D. Sven Lindmark, Gothenburg University

Master of Science in Management

Supervisor: Ph D. Luigi Marengo, LUISS Guido Carli

I

Technology Acceptance of Future Decision Makers

An investigation of Behavioral Intention Toward Using Artificial Intelligence Tools in Decision

Making

By Emelie Gunnarsson

© Emelie Gunnarsson

School of Business, Economics and Law, University of Gothenburg

Vasagatan 1 P.O. Box 600, SE 405 30 Gothenburg, Sweden

All rights Reserved.

No part of this thesis may be reproduced without the written permission by the author.

Contact: [email protected]

II

Acknowledgements

I would like to express my sincere gratitude to the people that contributed to this master thesis. Firstly,

to the professor and supervisor of the School of Business, Economics and Law, University of

Gothenburg, Sven Lindmark for his continuous support, guidance and motivation throughout the

research process. I would also like to express my appreciation to professor Luigi Marengo at LUISS

University, Rome, for his immense knowledge on the area of Decision Making and his valuable input

to this master thesis. Furthermore, I would like to express my gratitude to my partner John Skoglund,

who has been encouraging me and believed in me during this master thesis journey. At last, I would

like to express my gratitude to my dear friends Izabella Jedel and Angus Palmberg who has brought

in valuable feedback in the research process and who were motivating me and providing me positive

energy to complete this master thesis.

Gothenburg, June 5th, 2020

______________________

Emelie Gunnarsson

III

Abstract

Essay/Thesis: 30 ECTS

Program: MSc in Innovation and Industrial Management &

MSc in Management

Semester/Year: Spring/2020

Supervisor: Sven Lindmark & Luigi Marengo

Keywords: Technological advancements, Technology, Big data, Data driven

decision making, Decision Making (or DM), Decision Support System

(or DSS), Decision Support, Decision aid, Intelligent decision support

system, Artificial intelligence (or AI), Automation, Risk, Technology

Acceptance Model (or TAM), History, Development, Definition,

Concepts and Organization (or Organisation).

Background & Purpose: Rapid technological advancements makes it difficult for organizations

to keep up, this alongside with the world becoming more and more

digital results in large amount of data being produced. The

technological advancements and the large amount of data available

results in that AI can be used as a decision making tool in organizations,

enabling better decision making. This because humans are not rational

or logical when making decisions. The concept of bounded rationality

is introduced and it may be explained as that humans take shortcuts in

the thought process which result in suboptimal decision making. AI

enables data driven decisions to be taken and the restrictiveness of the

human mind indicates that the old way businesses made decisions are

about to change. This thesis focuses on the role of AI in decision

making. The Technological Acceptance Model (TAM) is a theory that

explains to what degree a user is intending to perform a certain

behavior. Because students are likely to reach a decision maker

position in the future working environment, the interest of this study is

to examine students’ behavioral intent to use AI as a decision support

system in their future workplace. This result in the following research

question: What factors in TAM influence business students’ behavioral

intent to use AI in decision making?

IV

Theory: The theory chapter is based upon four main sections; technological

Advancements impact on the business environment, decision making

theory, AI and TAM. The technological advancements is divided into

three subcategories; organization, people and data & information.

Decision making theory includes characteristics, definition and Simons

decision making process with five different phases; intelligent, design,

choice, implementation phase and lastly feedback. In the section about

AI, historical aspects are highlighted and a definition used in this report

is provided. Later, AI is connected to an organizational perspective and

linked to decision support system (DSS) and automation of decision

making. At last, the framework TAM is presented and its history

alongside a description of the components of TAM is provided.

Methodology: The methodology chapter is divided into four subcategorize; research

strategy, research design, research method and data analysis. Firstly,

the research strategy is described and this thesis is based upon a

deductive approach, whereas the assumptions objectivism and

positivism is the basis and this thesis takes upon a quantitative

approach. In the section research design, the quality criteria of this

thesis is described and the chosen research design cross-sectional is

described. Later, in the research method section, the secondary and

primary data collection is accounted for. At last, the data analysis is

motivated.

Result & Discussion: This study concluded that perceived usefulness (PU) is 2,43 times as

important than perceived ease of use (PEU) in influencing behavioral

intention (BI) among students of using AI as a decision making tool.

This information can be valuable when constructing systems because

it shows that time spent on improving PEU might be more useful for

the user and lead toward a stronger behavioral intent of using the

system if more time is spend on increasing the PEU of the system.

V

Table of Contents

1. Introduction 1

1.1 Background 1 1.2 Research Problem 2 1.3 Purpose and Research Question 4 1.4 Delimitations 5 1.5 Disposition 6

2. Theoretical Framework 7

2.1 Technological Advancements’ impact on the business environment 7

2.1.1 Organization 7 2.1.2 People 7 2.1.3 Data & Information 8

2.2 Decision Making Theory 8

2.2.1 Characteristics of Decision Making 8 2.2.2 Definition of Decision Making 9 2.2.3 Decision Style & Decision Makers 9

2.2.3.1 Decision Style 10 2.2.3.2 Decision Makers 10

2.2.4 Phases of the Decision Making Process 11

2.2.4.1 Intelligent Phase 11 2.2.4.2 Design Phase 11 2.2.4.3 Choice Phase 12 2.2.4.4 Implementation Phase 13 2.2.4.5 Feedback 13

2.3 Artificial Intelligence 13

2.3.1 History of Artificial Intelligence 13

2.3.1.1 1950s 14 2.3.1.2 1970s 14 2.3.1.3 1990s 14 2.3.1.4 2010s 15

2.3.2 Definition of Artificial Intelligence 15 2.3.3 The Fundamental Concepts of Artificial Intelligence 16 2.3.4 AI & Organization 18 2.3.5 Artificial Intelligence and application to Decision Making Theory 19

2.3.5.1 Decision Support System 19 2.3.5.2 The Level of Automation in Decision Making 20

2.3.6 Limitations of Artificial Intelligence 20

2.4 Technology Acceptance Model 21

2.4.1 History of the Technology Acceptance Model 21 2.4.2 The Constructs of the Technology Acceptance Model 23

2.4.2.1 Perceived Usefulness (PU) 23 2.4.2.2 Perceived Ease of Use (PEU) 23

VI

2.4.2.3 Perceived Enjoyment (PE) 24 2.4.2.4 Attitude toward using (A) 24 2.4.2.5 Behavioral Intent (BI) 24 2.4.2.6 Actual Use 24

2.4.3 Theoretical Foundations of Technological Acceptance Model 25

2.4.3.1 Self-Efficacy Theory 25 2.4.3.2 Motivational Theory 25 2.4.3.3 Cost-benefit paradigm, behavioral decision theory 25

3. Methodology 27

3.1 Research Strategy 27

3.1.1 Deductive Approach 27 3.1.2 Assumptions, Objectivism & Positivism 28 3.1.3 Quantitative Approach 29

3.2 Research Design 30

3.2.1 Quality Criteria 30

3.2.1.1 Reliability 30 3.2.1.2 Replicability 31 3.2.1.3 Validity 31

3.2.2 Cross-Sectional Research Design 32

3.3. Research Method 33

3.3.1 Secondary data collection 33

3.3.1.1 Databases 34 3.3.1.2 Key words 34

3.3.2 Primary data collection 34

3.3.2.1 Questionnaire Design 35 3.3.2.2 Sampling 36

3.4 Data Analysis 37

4. Empirical Findings 39

4.1 Internal Reliability Test, using Cronbach Alpha 39 4.2 Factor Analysis 40 4.3 Regression Test & Parameter Estimates 42

4.1.2 Causal diagram of Technology Acceptance Model 45

5. Discussion 47

5.1 Main Findings 47

5.1.1 Perceived Usefulness 48 5.1.2 Perceived Ease of Use 49 5.1.3 Attitude Toward Using 50 5.1.4 Behavioral Intent 51

5.2 Comparison to Davis Original Model 51 5.3 The Future Decision Environment in Organizations 52

6. Conclusion 53

6.1 Main Findings 53 6.2 Implications 53 6.3 Suggestion for Future Research 54

VII

7. Bibliography 55 8. Appendixes 60

A - Questionnaire Guide 60 B - Correlation Matrices 68

B.1 Perceived Ease of Use 68 B.2 Perceived Usefulness 68 B.3 Attitude Toward Using 69 B.4 Behavioral Intention 69

C- List of Exhibits 70

C.1 List of Figures 70 C.2 List of Tables 70 C.3 List of Abbreviations 71

1

1. Introduction

________________________________________________________________________________

This chapter aims to present an overview of the background setting, research problem, purpose and

research question. Furthermore, hypotheses are presented. Lastly, limitations of the thesis and a

depiction of the thesis disposition is provided in order to give the reader the structure of the thesis as

a whole.

________________________________________________________________________________

1.1 Background

Society and business environment are characterized by rapid technological advancements. The speed

and the advancements are progressing at such a level that it is becoming hard for organizations to

keep up. Nonetheless, it is of increasing importance that companies are able to manage the setting in

which businesses are currently exposed to.

“Companies don’t fail because of changes in the environment, they fail because their

leaders are either unwilling or incapable of dealing with said change”.

- (Boss, 2016)

Rosenberg (2014) emphasizes that one of the strongest drivers and creators of economic growth is

technological development. Furthermore, technological advances have the ability to cause disruptive

changes in the business environment (ibid) in which organizations operate. One example of a digital

disruptor is Artificial Intelligence. There is no clear definition of Artificial Intelligence that has been

widely accepted, due to the vast amount of applications of the broad range of technologies it provides.

However, in this report, the following definition will be used: “The theory and development of

computer systems able to perform tasks normally requiring human intelligence” (Oxford English

Dictionary, n.d.). Artificial Intelligence, or AI, interrupts the status quo, consequently impacting

businesses, companies, industries and professions (Doshi, Balasingam & Arumugam, 2020). Again,

adapting to change keeps us at the forefront of the competitive edge (Boss, 2016).

The world is becoming more and more digital, resulting in a large amount of produced data. This data

needs to be properly collected and managed (Pereira, Belo & Ravesteijn, 2018). Big data, in

comparison to data, is characterized by three main features that differentiate big data from data. At

first, huge amounts of data that are being produced (volume), that the data is generated at a higher

pace (velocity) and the sources from which the data is collected require different technologies

2

(variety) (ibid). Big data is an asset that enables enhanced insights, decision making and process

automation (Gartner IT Glossary, n.d.). Big data technology enables organizations to lower costs,

increases speed of processing information and allows new data and models that results in better

decision making (Davenport, 2014, p. 115).

Despite the advantages of AI as a decision making tool it might not be used to the expected extent.

This is because there is a discrepancy between normative decision theory, how humans ought to make

decisions, and descriptive decision theory, how decisions actually are made. Humans are not rational

or logical when making decisions. The base of neoclassical economic theory is that individuals will

rationally pursue behavior which minimizes the level of pain while maximizing the level of pleasure

(Swanson, 1996). Simon (1955) presented an alternative to mathematical modeling of decision

making; which is the start of the behavioral economic theory. Bounded rationality, the idea that

humans take shortcuts which will result in suboptimal decision making (ibid). In the late 1960s

cognitive psychology was added to economic theory by Kahneman and Tversky's contribution; brain

as an information processing device. Economist models of rational behavior were compared to

cognitive models of decision-making under risk and uncertainty (Kahneman & Tversky, 1979), so

called Prospect Theory. Another contribution to bounded rationality was “nudge”, referring to that

choice architectures can be altered to nudge humans into making better decisions without restricting

any option (Thaler, 2018). However, it has been argued that changing the choice architectures will

not help humans make long-term behavioral changes (Wright & Ginsberg, 2012).

1.2 Research Problem

From the rapid increases of technological advancements mentioned above, combined with the

restrictiveness of the human mind, it can be argued that the old way businesses made decisions are

about to change. The rest of the thesis therefore aims at investigating how decisions can be made

from technologies instead, and thus focuses on the role of Artificial Intelligence in decision making.

Technological advances, such as AI, develop at a rapid speed, and the number of companies that will

make use of the numerous possibilities AI provides increases. However, it is important to bear in

mind that AI does not only have positive implications but also entail risk. Amodei et al. (2016)

emphasizes the importance of understanding the challenges AI entails in order to be successful in

developing useful, relevant and important AI systems that will make organization thrive. The large

amount of data available for managers has changed how decisions are made, less on intuition and

more based on data (Brynjolfsson & McElheran, 2016).

3

Are humans the best decision makers in a business environment? According to Kahneman,

Rosenfield, Gandhi and Blaser (2016) humans are unreliable as decision makers, because the human

mind is influenced by highly irrelevant factors to make a corporate decision, such as hunger, mood

or today's weather. A study was conducted and the result was algorithms, one form of AI that is called

machine learning, was more accurate in half of the studies, and had the same accuracy as humans in

the remaining studies (ibid). Due to cost efficiencies related to machine learning compared to human

capital, algorithms also won that battle of accuracy (ibid). Marwala and Hurwitz (2017, p. 172) also

argue that decisions would be more reliable if they were made by an AI machine rather than humans.

One of the reasons mentioned is that intelligent machines reduce the impact of bounded rationality in

decision making when a machine rather than a human makes a decision (ibid, p. 41). Another

argument is that machines reduce the degree of information asymmetry in the market, resulting in

more rational markets and improve decision making ability (ibid, p. 105).

“Studies show that algorithms do better than humans in the role of decision maker”.

- (Kahneman et al., 2016)

What will the future decision environment look like? Von Krogh (2018) argues that Artificial

Intelligence will not substitute humans, rather work as an augment in order to make better decisions

that are to a higher degree based on data, while using the machines to process large amounts of data.

In contrast, Agrawal, Gans and Goldfarb (2019) believe that AI tools will work both for automating

decisions as well as an instrument to improve decision making made by humans. Most likely, the

decision environment will change in the future, and decisions made by humans, with the assistance

of machines, or decisions made in full by machines will increase (Marwala & Hurwitz, 2017, p. 7).

Furthermore, technologies such as AI have led to major changes in the economic theories (ibid, p.

103).

That the future holds more AI in managerial decision making is without a doubt. Are corporations

and future decision makers ready to make the switch to more AI as a decision making tool? Because

of the rapid advancements in AI, it has the possibility to play a bigger role in the society. Research

on innovation management has shown that in order for the technology to take on this role, it has to

be accepted by the users. One method to examine whether the future decision makers are ready to

increase the level of AI as a decision making tool is to use the Technology Acceptance Model, TAM

to examine acceptance behavior.

4

TAM is grounded in the Theory of Reasoned Action, a theory that explains to what degree a user is

intending to perform a certain behavior (Djamasbi, Strong & Dishaw, 2010). In order to find the

behavioral intent, BI, of an individual it is divided in different aspects; Perceived Usefulness (PU),

Perceived Ease of Use (PEU) whereas PEU influences PU (ibid). TAM is one of the most influential

Information Systems theories (ibid). This model will be used in this study to understand the

behavioral intent of future decision makers, within the business environment, to use Artificial

Intelligence as a tool in the decision making process.

Students undertaking a bachelor’s or a master’s degree within the field of business and economics

are likely to reach a decision maker position in the future working environment. Consequently, this

study will examine students' behavioral intent to use Artificial Intelligence as a decision aid in their

future workplace.

“The impact of a technological innovation will generally depend not only on its inventors,

but also on the creativity of the eventual users of the new technology. “

- (Rosenberg, 2004, p. 5).

1.3 Purpose and Research Question

The purpose of this thesis is to examine the behavioral intent of future decision makers in the use of

Artificial Intelligence as a decision making tool, this by using the Technology Acceptance Model.

The study focuses on students in Sweden, within the field of business and management. The research

question is displayed below:

What factors in TAM influence business students’ behavioral intent to use

Artificial Intelligence in decision making?

This research question will be broken down into a few hypotheses in order to make sense of the

interpretation of the result. Hypothesis one to five is to test if the model holds true in this survey data,

this is necessary in order to derive hypothesis six, which is related to the research question and in

detail, determine the relative effect PU and PEU has on BI. Davis (1985) concluded all of these

hypotheses to be true in his determination of TAM. Hypothesis four, however was formulated

differently in his study (have a significant effect), however because in the result Davis proved it

falsified, I changed hypothesis four to (have not a significant effect) in order to be in line with the

model constructed from his research.

5

H1: Perceived ease of use (PEU) will have a significant effect on perceived usefulness (PU).

H2: Perceived usefulness (PU) will have a significant effect on attitude toward using (A),

controlling for perceived ease of use (PEU).

H3: Perceived ease of use (PEU) will have a significant effect on attitude toward using (A),

controlling for perceived usefulness (PU).

H4: Attitude toward using (A) will not have a significant effect on behavioral intent (BI),

controlling for perceived usefulness (PU).

H5: Perceived usefulness (PU) will have a significant effect on behavioral intent (BI),

controlling for attitude toward using (A).

The result of Davis (1985) study showed that perceived usefulness had a greater impact on attitude

toward using and consequently behavioral intent than perceived ease of use. Consequently, this study

wants to test if that relationship also holds true in this survey data.

H6: Perceived usefulness (PU) will have a greater impact on behavioral intention (BI) than

perceived ease of use (PEU).

1.4 Delimitations

This paper aims to provide the reader with an understanding of the possible impacts of Artificial

Intelligence to the decision theory and the person in the decision maker position, from a managerial

standpoint. For this reason, the technological or technical aspects of Artificial Intelligence are

disregarded. Furthermore, ethical dilemmas considering the implementation of Artificial Intelligence,

such as fear about if machines will outrun humans, or any cases of a dystopian future due to Artificial

Intelligence will not be discussed.

The participants of the surveys’, that were conducted, consisted of students that are pursuing a

bachelors’ or masters’ degree at School of Business Economics and Law, University of Gothenburg.

This study is limited to understand the future decision makers of the Swedish business market and

their acceptance to AI as a decision making tool. Hence, this study is focusing on the future decision

makers of the Swedish business market.

6

1.5 Disposition

This thesis is divided into six main sections, introduction, theoretical framework, methodology,

empirical findings, analysis and discussion and at last conclusions. The main idea of each chapter is

provided in figure 1.1 below.

Figure 1.1 Overview of Research Process

7

2. Theoretical Framework

________________________________________________________________________________

This chapter presents the theoretical foundation of this thesis. First, an introduction to the

technological advancements’ impact on the business environment will be provided, divided into three

parts, organization, people and data and information. Second, a comprehensive presentation of

decision making theories based upon Simons theory of the decision making process will be presented.

Third, an overview of AI, its definition, development and progress toward taking on a bigger role in

the society will be presented. At last, a comprehensive presentation of the framework TAM which is

the basis of quantitative analysis is carried out in relation to.

________________________________________________________________________________

2.1 Technological Advancements’ impact on the business environment

2.1.1 Organization

The technological advancements taking place all over the globe is affecting the business environment

and changing it. According to Cukier (2019) major technological changes happened since 2008,

particularly in the areas of communications and AI. These technological changes have led to a new

business landscape for companies. According to Ahlstrom (2010) the technological changes currently

taking place are expanding industry boundaries, thus enabling greater competition and coopetition.

Because of globalization and technological advancements, which enables greater competition and

coopetition, companies need to be able to adapt to the changing landscape in order to survive. Being

incapable of dealing with the change taking place because of technological advancements’, might

lead to that a company will cease to exist (Boss, 2016). The speed of the changing taking place is

rapid, consequently making it difficult for companies to keep up. Nonetheless, it is crucial for a

company’s survival to be able to keep up with the changes that affect the business environment to

stay relevant (ibid).

2.1.2 People

The people in the organization are very important to determine if the technological advancement will

be accepted (Davis, 1989; Davis, Bagozzi & Warshaw, 1989), hence used or not. That being said, not

all employees in the organization are willing to adapt to change, consequently it is important to have

the right personnel (Rosenberg, 2004, p. 5). Personal that are willing to strive for change in order to

make the development of technology useful. Because the technology on its own does not have a direct

8

value, it has to be accepted and used by the personal throughout the organization in order for the value

of the technology to be established. Consequently, having the right personnel is important when new

technology has to be managed in order to adapt to changes taking place in the business environment.

Decision Support System, DSS, is a technological advancement that can help people at decision

making positions in organizations to make better decisions. Humans are not logical or rational when

making decisions. This has been shown in numerous studies (i.e. Simon, 1955; Camerer, 1998), there

is a discrepancy present between how humans ought to make decisions and how decisions actually

are made. Because of this discrepancy, humans are unreliable as decision makers (Kahneman et al.,

2016). DSS is part of the field of Artificial Intelligence, hence the focus forward will be to investigate

the role of Artificial Intelligence in decision making.

2.1.3 Data & Information

One consequence of the digitalization taking place in the society and the business environment is that

the number of data both produced and collected is increasing rapidly. Pereira, Belo and Ravesteijn

(2018) emphasizes the importance that the data needs to be properly collected and managed. This is

because data contains a lot of numbers that can be translated into information and this information is

valuable for companies because it can be used as insights when making decisions.

Big data is differentiated from data, by the description of these three features; volume, velocity and

variety (ibid). Because big data contain a lot of sequences of information while being able to

reproduce at such rapid speed big data is a tool that may favor the decision making process by

enhanced insights being present (Gartner IT Glossary, n.d.). Davenport (2014, p. 115) emphasizes the

fact that the technology advancements that have taken place, that enables big data to be collected and

managed, has a potential to lower organizational costs and increase the speed of processing

information. All in all, the big data technology enables new data and models to arise which will result

in better decision making in organizations (ibid, p. 115).

2.2 Decision Making Theory

2.2.1 Characteristics of Decision Making

Decision making may be influenced by several factors. Some of them will be discussed in the below

section. One of the characteristics of decision making is groupthink, referring to the process when

bad decisions are taken by a group because its members do not express opinions or suggest new ideas.

9

Decision makers may be interested in evaluating what-if scenarios, alternatively experimentation with

a real system that might result in failure (Sharda, Delen & Turban, 2018, p. 70).

Changes to the environment in which decisions are taken may occur continuously without the

awareness of the decision makers, thus might lead to invalidating assumptions about a situation (ibid,

p. 70). Another effect of changes in the decision environment is that the quality of the decision might

reduce because time pressure is put on the decision maker (ibid, p. 70).

When it comes to the amount of data and information available it can cause problems in decision

making because collecting information and analyzing a problem takes valuable time from the

managers, and it is hard to detect when to stop the analysis phase and actually take the decision (ibid,

p. 70). Another implication in relation to data and information might be that there is not sufficient

information in order to take an intelligent decision (ibid, p. 70). At last, there is a risk of information

overload, so to say, too much information available (ibid, p. 70).

2.2.2 Definition of Decision Making

Decision making is often referred to the process of choosing among different alternatives of actions

in order to achieve one or more goals. According to Simon (1978), managerial decision making

involves the entire management process. Simon exemplifies this by talking about planning. There are

several decisions in relation to planning: What should be done? When? Where? Why? How? By

whom? Managers in organizations set goals or plan, and as previously made clear, planning involves

decision making (ibid). Organizing or controlling is two examples of other managerial functions

which also involves decision making (ibid).

Sharda, Delen and Turban (2018, p. 71) emphasizes that efficiency often is not the main focus in

decision making tools, rather it may be a by-product obtained when pursuing the main goal

effectiveness. In this thesis, the above approach to decision making, including the entire management

process, will be followed due to it well aligned with the business area approach of this thesis.

2.2.3 Decision Style & Decision Makers

Decision aid is not useful for the end user unless it is adapted to appropriate methodologies in order

to work efficiently and correctly in the environment in which it operates. It is important to understand

how decision makers make decisions, to construct the appropriate decision tool. In order to

understand how people make decisions, the process and issues involved in decision making is

10

important. The focus on the decision making theory in this thesis will be on the business approach,

so to say, managerial decision making.

2.2.3.1 Decision Style

According to Sharda, Delen and Turban (2018, pp. 71-72) decision style may be described as how

decision makers think and react to problems that emerge. This can be decomposed as how they

perceive a problem, their cognitive responses and how values and beliefs vary between individuals

and situations (ibid, pp. 71-72). People go about making decisions in different ways, the path is not

always linear nor all steps of the general decision making process sequence followed (ibid, pp. 71-

72).

There are numerous different decision making styles. As stated above in section 2.2.3, it is important

that the system supports the manager. Hence, the fit between the manager, in detail the fit with the

decision style and the decision situation, is of high importance. As a consequence, the system should

be flexible and adaptable to different users (ibid, pp. 71-72). The system should augment the decision

makers' development of their own style, skills and knowledge (ibid, pp. 71-72). Identifying a decision

style can be done by different tests available, both o personality and temperament. By identifying the

person’s style can impose the most effective communication to take place and an ideal task can be

given for the best suited individual (ibid, pp. 71-72).

2.2.3.2 Decision Makers

Decisions in organizations are often done by individuals in smaller organizations and done by groups

in medium to large organizations. It is important to be aware of difficulties that can arise in

organizations when making decisions as a group (ibid, p. 72). The group may be composed of

personnel from different departments having different interests in the matter etc. The individuals in

the group may also have different decision styles and personality types that makes it harder to reach

an effective decision with the best possible outcome.

But what is the best possible outcome? For whom? A decision support system may help determine

different alternatives when conducting a decision, there might be group thinking present and other

biases might influence the quality of the decision. A decision aid may possibly enhance the quality

of the decision made. But in order for a decision to be fully automated, a human decision maker must

decide such a thing (ibid, p. 72). Hence, human decision making in organization is of high

importance.

11

2.2.4 Phases of the Decision Making Process

Simon (1960, p. 2) defined different phases in the decision making process; intelligence, design,

choice, implementation and feedback, whereas the implementation phase was a later addition.

2.2.4.1 Intelligent Phase

According to Simon (1960, p. 2) the intelligence phase is the first phase of decision making, the

decision maker examines reality while identifying and defining the problem. The intelligence phase

involves activities that are aiming at identifying problem situations or opportunities (Sharda, Delen

& Turban, 2018, pp. 74-76). After the organization's goals and objectives related to an issue of

concern is set, a determination takes place whether they are being met or not (ibid, pp. 74-76). When

the issue is determined, it is important to determine the significance of the problem and make a proper

definition of it (ibid, pp. 74-76).

The classification of the issue is important because it is possible to place it in a definable category,

which can lead to a solution approach (ibid, pp. 74-76). However, problems in organization might be

of complex nature. Hence it is important to be aware that complex problems might be divided into

subproblems. One of the most important decisions in this phase is related to problem ownership.

Problem ownership refers to assigning authority to solve the problem (ibid, pp. 74-76). Before moving

on to the design phase, this phase often ends with a formal problem statement (ibid, pp. 74-76). This

stage may also include monitoring the results of the implementation phase.

2.2.4.2 Design Phase

In this second phase, a model representing the system should be constructed (Simon, 1960, p. 2). The

model is constructed by making assumptions about the reality and by writing down relationships

among the variables (Sharda, Delen & Turban, 2018, p. 73). Later, all alternative courses of actions

are identified and evaluated for feasibility.

In the design phase is often a model used because it is a simplified representation of reality, this

because reality often is too complex. Later, the selection of a principle of choice, a description of the

acceptability of the solution approach, is set (ibid, pp. 77-84). The two most common principles of

choice are normative and descriptive.

A normative model can be described as a model in which the chosen alternative is the best of all

possible alternatives (ibid, pp. 77-84). Normative decision theory is based upon humans wanting to

maximizing the attainment of goals, the decision maker is rational, all possible viable actions are

12

known and decision makers can rank their desirability of all actions (ibid, pp. 77-84). However, as

discussed in the introduction, decision makers are not rational hence the introduction of descriptive

models.

A descriptive model is often mathematically based and describes things as they are believed to be

(ibid, pp. 77-84). The descriptive approach normally checks the performance only for a given set of

alternatives, rather than for all alternatives in the normative model (ibid, pp. 77-84). This is resulting

in that the alternative selected often is of satisfactory level rather than optimal. The most common

approach in descriptive modeling methods is simulation. According to Simon (1960, p. 6-7) most

human decision making involves a willingness to settle for a satisfactory solution. This because there

often is time pressure or that the ability to achieve optimization is too time consuming that it might

not be worth it due to the marginal cost to obtain it. Simons (1955) theory of bounded rationality,

humans have a limited capacity for rational thinking, is one explanation of why many models are

descriptive rather than normative. An experienced decision maker often recognizes the trade-off

between the marginal cost of obtaining further information versus making a better decision (Simon,

1960, p. 6-7).

A common model used when making decisions is scenarios. According to Sharda, Delen and Turban

(2018, pp. 77-84) a scenario may be described as a statement of assumptions about the environment

of a particular system in a given time period. The scenario describes the decision and different

variables and parameters for a specific situation (ibid, pp. 77-84). What is commonly used in business

practices is a description of possible scenarios. There might be thousands of scenarios for every

decision, but some are more used than others. Commonly adopted in businesses is the worst possible

scenario, the best possible scenario or the most likely scenario.

2.2.4.3 Choice Phase

The choice phase includes proposed solutions to the model, not necessarily to the problem it

represents (Sharda, Delen & Turban, 2018, p. 73). The solution has to be tested and the result must

be of satisfactory level and show viability. This phase is often perceived as the most critical in

decision making (ibid, p. 85). In this phase, is the decision made and the commitment to follow actions

in order to reach the desired outcome. There might be some difficulties in detecting the boundary of

the design and the choice phase because some activities can be related to both phases (ibid, p. 85). In

order to have a successful phase it is important to find an appropriate solution to a model. However,

this is not the same as having a recommended solution successfully implemented.

13

Two models often used in order to derive at different choice paths is sensitivity analysis and what-if

analysis (ibid, p. 85). Sensitivity analysis is used to determine the robustness of the alternatives and

what-if analysis is used to explore any larger changes in the parameters (ibid, p. 85).

2.2.4.4 Implementation Phase

Implementation is perceived as the fourth phase of decision making, however this was a later addition

made by Simon, not part of the original model. This phase is about implementing the decision,

whereas successful implementation results in solving the problem and failure leads to a return into an

earlier phase in the process (Sharda, Delen & Turban, 2018, p. 73). The implementation can be seen

as the initiation of a new order of things or the introduction of change (ibid, pp. 85-86), not to forget

that change needs to be managed.

The implementation phase is complex because this is most often a long, involved process in putting

a recommended solution to work (ibid, pp. 85-86). There are also often challenges present when

implementing the decision, such as: resistance to change in the organization, the degree of support of

top management or training.

2.2.4.5 Feedback

In all of the phases mentioned above: intelligent, design, choice and implementation, is it possible to

return to an earlier phase because of the feedback loops available. Consequently, feedback is not seen

as a fifth phase, rather an iterative process whereas feedback may proceed and that can move you

back to a previous step in the decision making model. However, the feedback is most often done after

the implementation phase in order to learn from previous decisions and improve before the next (ibid,

p. 86). Hence, it is important to employ analytics also in the feedback process and not only when

identifying the problem or solution (ibid, p. 86).

2.3 Artificial Intelligence

2.3.1 History of Artificial Intelligence

Even if AI was established as an academic discipline back in the 1950s, the area was receiving limited

practical interest for over half a century (Haenlein & Kaplan, 2019). Because of the recent

improvements made in Big Data and computing power the interest of AI has reached the public

conversation (ibid).

14

2.3.1.1 1950s

During the second world war, Alan Turing developed a code breaking machine called The Bombe for

the British government, with the purpose of deciphering the enigma code used by German soldiers.

The Bombe has later been considered the first working electro-mechanical computer. In 1950, Turing

published “Computing Machinery and Intelligence” which explained how to create intelligent

machines and how to test their intelligence, the so-called Turing Test. The Turing Test can be

described as if a human is interacting with a machine and with another human, and cannot distinguish

them apart, the machine is said to be intelligent.

In 1952, a machine that was able to learn how to play checkers was created by Arthur Samuel,

researcher at IBM. The machine was provided with insights of human experts at the games and it

plays against itself in order to learn to distinguish good moves from bad. This is marked as a milestone

in the history of AI because the machine's ability to learn strategies by playing against itself (Royal

Society, 2017, pp. 26-28).

The Dartmouth Workshop took place during the summer of 1956, and at this workshop the term

Artificial Intelligence was coined by John McCarthy and Marvin Minsky. Participants that took part

in this workshop was later to be considered as the founding fathers of AI, i.e. Nathaniel Rochester

and Claude Shannon (Haenlein & Kaplan, 2019).

2.3.1.2 1970s

Several success stories in the field of AI in the 60s, led to funding being given to AI research, thus

more projects were enabled in the field of AI. However, in 1973 the U.S Congress started to criticize

the spending in AI research and they questioned the outlook of AI presented by researchers. The

Lighthill report noted that “in no part of the field have the discoveries made so far produced the major

impact that was then promised”. As a result, unfortunately both the British and U.S government

stopped the funding of AI projects (Haenlein & Kaplan, 2019; Royal Society, 2017, p. 28).

2.3.1.3 1990s

The machines created earlier had not been excellent just a bit higher than the average human in

performance. Not until 1997, when the first computer chess-playing system that was able to beat the

reigning world chess champion. This machine, compared to the earlier version could exploit the

increased computing power available in 1990s in order to perform searches of potential moves at a

larger scale, process over 200 million moves per second, and then just choose the best one (Campbell,

15

Hoane & Hsu, 2002). James Lighthill was at the end proven wrong, almost 25 years later, when the

machine beat the world champion present at the time (Haenlein & Kaplan, 2019).

2.3.1.4 2010s

In 2011, IBM’s robot Watson competed in Jeopardy and he was up against two earlier champions at

the gameshow and Watson won (Royal Society, 2017, p. 28).

Later, in 2016 AlphaGo, created by Google DeepMind, won four out of five games of Go against Lee

Sedol, who has been the champion at the game for more than a decade at that time (ibid). Go can be

described as substantially more complex than chess and it has long been perceived that computers

could not beat humans in this game. AlphaGo, was able to beat the champion because it was

developed by a specific type of neural network, so called deep learning (Silver et al., 2016).

2.3.2 Definition of Artificial Intelligence

The coined term Artificial Intelligence lacks a broadly agreed definition. One reason for this is that

AI is a very broad concept that covers a range of technologies and vast amounts of areas resulting in

it is hard to precise and distinct a definition (McKinsey, 2017). Burgess (2018, p. 5) highlights another

reason why AI is hard to define, there is a lack of general theory of how and what “intelligence”

consists of. A third reason why a definition of AI is hard to determine, is because its nature, what was

considered intelligent yesterday is standard today and consequently no longer perceived as intelligent

to the same extent (McKinsey, 2017; Capgemini, 2017).

Despite the fact that it is hard to establish a specific definition of AI, most of the suggested definitions

have some elements in common. These definitions relate to systems that think like humans or act like

humans and either think rationally or may act rationally. The definition that will be used throughout

this report is:

“The theory and development of computer systems able to perform

tasks normally requiring human intelligence”

- (Oxford English Dictionary, n.d.)

This definition fulfills the common agreement that the definition should constitute of the system being

able to act like humans and thinks rationally.

16

2.3.3 The Fundamental Concepts of Artificial Intelligence

AI is a widespread term that contains many different technologies and concepts. Due to the nature of

this thesis, the focus on the technological approach of AI will not be accounted for. Hence, the focus

on this short explanation of the different concepts of AI will have a business view of it, and details of

how AI works will not be discussed going forward. The most important trait of AI is that it can learn

and evolve based on its own experiences (Burgess, 2018, p. 2). Burgess (2018, p. 2) emphasizes that

it is important that AI is expressed in simple-enough terms in order to businesses to take on AI and

understand how it may help their business. If the use of AI is not communicated correctly there is an

underlying risk that AI never will reach its full potential. Table 2.1 below will explain the most

common concepts of AI.

17

Concepts Explanation of Concepts

Machine Learning (ML) The process by which an AI uses algorithms to perform AI functions.

Artificial Neural

Network (ANN)

Created when wanting AI to improve in a certain action. ANN are

designed to be similar to the human nervous system and brain. It gives

AI the ability to solve complex problems by breaking different stages

of learning into levels of data.

Deep Learning (DL) The result of when ANN gets to work. The layers in ANN processes

data and the AI gains a basic understanding. DL refers to the action of

learning “why” instead of understanding what something is.

Supervised Learning

(SL)

One method to use when training an AI model. By providing the

machine with the correct answer ahead of time, the AI knows the

answer and the question. The most common method of training

because it yields the most data.

If you want to understand why or how something happens, an AI can

look at the data and determine connections using a SL method.

Unsupervised Learning

(UL)

In contrast to SL, in the method of unsupervised learning, the AI is not

given the answer to a question. The machine is feed with data and it is

allowed to find whatever patterns it discovers.

Algorithm Algorithms can be described in AI as rules that teach computers how

to figure things out on their own. It is a construct of numbers and

commands, and they are proven very useful.

Black Box Black box learning refers to the fact that AI often requires a lot of

complex math that often cannot be understood by humans, despite this,

the output is useful information. This concept is called black box

learning, we do not care of how the computer arrived at the decision,

because we know what rules it used to get to that point.

Table 2.1 Concepts of Artificial Intelligence (Greene, 2017).

18

2.3.4 AI & Organization

One question often debated is whether the development and research done in AI will result in people

losing their jobs to robots. This issue has been compared to the automation of the manufacturing

process (Haenlein & Kaplan, 2019). The automation of the manufacturing process resulted in the loss

of blue-collar jobs, and the development of AI might result in less need for white collar employees

even in high-qualified professional jobs (ibid).

Burgess (2018, pp. 6-8) argues that AI will have a severe impact on jobs, but AI can also augment

the amount of work done because of its effectiveness. AI provides the opportunity that the personal

can use their time to properly analyze the information that AI has delivered, instead of spending hours

collecting data and researching cases (ibid, pp. 6-8).

The question of debate here is whether the net impact on work will be positive or negative (ibid, pp.

6-8). Will automation with the help of AI create more jobs than it destroys? However, it may be

argued that the activities that humans will perform in the organization might be more enjoyable and

the augment of AI in the organization might result in performance gains that might possibly outweigh

the loss of jobs that have been replaced by AI. It is clear that AI is a huge disruptor and will affect

several aspects in our lives (ibid, pp. 6-8). Similarly as with the automation process in manufacturing,

we will be able to adapt to this new way of working, but there might be a painful transition period

(ibid, pp. 6-8).

According to Haenlein and Kaplan (2019) one way to avoid unemployment at a larger scale, caused

by the improvement in AI and the adoption of such in organizations, is regulation. One suggested

regulation is that companies are required to spend a certain percentage of the money saved because

of automation in training personnel for other jobs which cannot be automated (ibid). Another

suggestion is that states may decide to limit the use of automation (ibid). A third suggestion is that

companies might restrict the number of hours worked per day in order to distribute the work more

evenly across the employees (ibid). All in all, there are different regulative measures that can be taken

in order to reduce the unemployment rate because of the automation process of AI in organizations.

AI will become as much a part of everyday life, just like the internet or social media has done in the

past. Hence, AI will not only impact our personal life, but also transform how firms take decisions

and interact with stakeholders, such as employees and customers (ibid).

19

2.3.5 Artificial Intelligence and application to Decision Making Theory

2.3.5.1 Decision Support System

An early framework for computerized decision support was proposed by Gorry and Scott-Morton in

the early 1970s. This framework came to evolve into a new technology, namely decision support

system, or in short DSS. The base of the model is the two categorizes “type of decision” and “type of

control (Gorry & Scott-Morton, 1971). The type of decision can either be structured, semi structured

or unstructured and the type of control can be operational, managerial or strategic (ibid).

Similarly, as the definition of AI, there is no universally accepted definition of DSS. Different persons

refer to it differently from the next. What most of the definitions have in common is that it is used for

improving the quality of the decisions made by managers (Gorry & Scott-Morton, 1989; Sharda,

Delen & Turban, 2014, p. 43). In general, DSS uses data, models and knowledge to find solutions for

semi structured and unstructured problems (Sharda, Delen & Turban, 2014, p. 62). In this thesis DSS

will refer to a narrow view, a decision support application used in organizations to guide decisions.

Furthermore, this thesis will not focus on the components of common DSS because of its business

approach.

Sharda, Delen and Turban (2014, p. 70) emphasizes that in order to develop DSS that is useful for

companies there is a need to understand the important issues involved in decision making. The

framework of decision making developed by Simon (1960) can be connected to DSS whereas it can

help in the different phases. In the first phases, the intelligence phase, DSS has proven useful when

dealing with high levels of unstructured ness in the problems (Sharda, Delen & Turban, 2014, pp. 87-

88). The design phase is about generating and forecasting consequences for different alternatives, and

DSS is useful in this phase, i.e. financial and forecasting models etc., because most DSS have

quantitative analysis capabilities (ibid, pp. 87-88). In the third phase, the choice phase, DSS has often

been supported by what-if analysis and goal-seeking analysis, this because several scenarios may be

explored before deriving at a decision (ibid, pp. 87-88). In the last phase, the implementation phase,

DSS may be used when communicating decision explanation and justification (ibid, pp. 87-88).

According to Kacprzyk, Valencia-García, Paredes-Valverde, Salas-Zárate & Alor-Hernández (2018,

p. v) could a proper application of decision making tools increase productivity, efficiency and

effectiveness. Which is important in order for the business to keep an advantage over the competitors.

DSS will allow the decision makers to make optimal choices for technological processes, planning,

logistics or investments (ibid, p. v).

20

2.3.5.2 The Level of Automation in Decision Making

The question whether AI will play a role in decision making or not is clear. However, which role it

will play and how humans and AI systems may peacefully coexist is more important to discuss.

Haenlein and Kaplan (2019) emphasizes that there is a need for a distinction in companies which

decisions should be taken by humans, AI or in collaboration between the two. Shrestha, Ben-

Menahem and Von Krogh (2019) developed a framework on this matter in an attempt to explain under

which conditions organizational decision making should be made by humans, fully delegated to AI,

a hybrid or aggregated. Hybrid decision making refers to whether AI can be used as an input in human

decision making alternatively that humans decisions as an input to AI systems. Aggregated decision

making refers to the fact that humans and AI are making decisions in parallel with the optimal

decision determined by voting.

What will be the level of automation in decision making in organizations in the future? According to

Von Krogh (2018) Artificial Intelligence will not substitute humans, rather work as an augment in

order to make decisions to a higher extent data based. The machines have the ability to process large

amounts of data so why not make use of the data that are available when making decisions in

organizations (ibid). Agrawal, Gans and Goldfarb (2019) has a similar perception as Von Krogh, they

believe that AI will improve decision making made by humans but make the addition that some

decisions probably will be suitable for automating by AI. Marwala and Hurwitz (2017, p. 7) is certain

that the decision environment will change in the future, and follows Von Kroghs reasoning that

decisions made by humans with assistance of machines or automated decisions will increase in the

future.

All of these authors mentioned above are arguing that a change in how companies will make decisions

in the future will change due to the progress made in AI. To what extent automation of decision

making in organization can only future tell.

2.3.6 Limitations of Artificial Intelligence

There are numerous opportunities for companies to make use of technological advances, such as AI.

However, it is important to remember that AI not only has positive implications for companies, it also

entails risk. Amodei et al. (2016) emphasizes that in order to be successful in developing useful,

relevant and important AI systems it is important to understand the challenges of AI. Hence, a shorter

section describing limitations of AI follows.

21

AI is in its essence objective and without prejudice, however it does not refer to that systems based

on AI cannot be biased (Haenlein & Kaplan, 2019). All data put into training the AI system and any

biases present in the input data may persist and even be amplified (ibid). How can such errors in the

input data be avoided in order to reduce biased AI systems? Haenlein and Kaplan (2019) suggests

that the most probable solution is to develop a commonly accepted requirement regarding testing and

training of AI algorithms. This can be done in combination with some form of warranty, similar to

consumer protocols used for physical products (ibid). This solution would allow stable regulation

even if the technical aspects of AI systems evolve over time (ibid).

In reference to the biases discussed above, understanding the intent of humans is complexed. Current

ML methods have a limited understanding of humans, restricted to particular domains (Royal Society,

2017, p. 30). This could possibly present some challenges in collaborative environments when robots

are collaborating with humans or when the level of automation collaborates with humans, such as

driverless cars (ibid, p. 30).

Providing large amounts of labelled training data, which some approaches to ML requires, can be

time-consuming and resource-intensive (ibid, p. 30). Another difficulty with ML methods is that there

are many constraints in the real world that it is not straightforward how to include these constraints

with the ML method (ibid, p. 30). Being able to encode such constraints, would allow more data

efficiency in the learning process.

2.4 Technology Acceptance Model

2.4.1 History of the Technology Acceptance Model

The Technology Acceptance Model, or hereafter called TAM, emerged from the Theory of Reasoned

Action (TRA) developed by Fishbein and Ajzen in 1975. TRA was used in order to predict and

explain human behavior in different settings (Davis, Bagozzi & Warshaw, 1989). The idea of the

model was that a person's behavioral intention (BI) to perform a specific behavior, determines the

performance. The behavioral intention is determined by both a person’s attitude (A) and subjective

norm (SN) with relative weights most likely to be estimated by regression (ibid). Figure 2.1 displays

the structure of the TRA model.

22

Figure 2.1 Theory of Reasoned Action, TRA (Davis, Bagozzi & Warshaw, 1989, p. 984).

Davis (1986) was interested in the field of Management Information Systems (MIS). Consequently,

he wanted to develop a model to empirically test new end-users within information systems, called

the Technology Acceptance Model, (TAM). TAM was derived from the TRA model, and the major

difference between the two models is that TAM is specifically tailored for modeling user acceptance

within information systems (Davis, 1989; Davis, Bagozzi & Warshaw, 1989).

TAM was supposed to be a helpful model, not only for prediction of behavior but furthermore, also

for explanation (Davis, 1989). The explanation part of the goal is important, because when

practitioners can identify why a system may be unacceptable, corrective steps may be taken in order

to correct and make the system better. An important note of the model is consequently, to trace the

impact of external factors on internal beliefs, such as attitudes and intentions (Davis, Bagozzi &

Warshaw, 1989).

According to Davis (1989), there are two main beliefs that can explain user acceptance, perceived

usefulness and perceived ease of use. These beliefs are linked to attitudes and usage. Perceived

usefulness is defined as “the degree to which a person believes that using a particular system would

enhance his or her job performance” (ibid, p. 320) and perceived ease of use is defined as “the degree

to which a person believes that using a particular system would be free of effort” (ibid, p. 320). The

model will be explained in more detail in the following section “2.5.2 The Constructs of the

Technology Acceptance Model”.

Later, in 1992, an addition was made to TAM in terms of a third belief, perceived enjoyment. Earlier

research had shown (Davis, 1989; Davis Bagozzi & Warshaw, 1989) that perceived usefulness had

strong links to usage intentions, however the role of enjoyment in workplace computing had not been

23

proper examined at the time. Perceived enjoyment was defined by Davis, Bagozzi and Warshaw

(1992, p. 1113) as “the extent to which the activity of using the computer is perceived to be enjoyable

in its own right, apart from any performance consequences that may be anticipated”.

2.4.2 The Constructs of the Technology Acceptance Model

The TAM model illustrates user’s behavior by explaining two general elements of acceptance, or so

called beliefs, Perceived Usefulness (PU) and Perceived Ease of Use (PEU). Easy explained, the

distinction between PU and PEU can be compared with the distinction between subjective decision

making performance and effort (Davis, 1989). The model is visualized below, see figure 2.2.

Figure 2.2 Technology Acceptance Model, TAM (Davis, Bagozzi & Warshaw, 1989, p. 985).

2.4.2.1 Perceived Usefulness (PU)

Perceived usefulness is one of the main beliefs in this model and it can be explained as to what extent

the users believe that the way using new technology enhances the performance of their jobs (Davis,

Bagozzi & Warshaw, 1989). The idea of this belief is that if people believe that an application or a

technological advancement will help in order to enhance their job performance they tend to use it. As

the model implies, see figure 2.2, PU may be affected by both perceived ease of use and external

variables.

2.4.2.2 Perceived Ease of Use (PEU)

The idea of this second belief is that even though the application might be useful, if the system is too

hard to use there is a risk that the performance benefits are outweighed by the effort of using the

application (ibid). The easier the application is perceived in relation to another application, the more

likely it is to be accepted by users, all other things equal. PEU, as figure 2.2 illustrates, is to be

determined by external factors.

24

2.4.2.3 Perceived Enjoyment (PE)

This is the third belief of the model, an addition made in 1992, consequently not part of figure 2.2

that was created in 1989. Davis, Bagozzi and Warshaw (1992) wanted to investigate whether people

tend to use computers more at work because it is useful or because it is enjoyable, apart from the

anticipated increase in performance. Carroll and Thomas (1988) suggested that enjoyment may have

a strong linkage to user acceptance, and gives an example in relation to that in the form of the Apple

Lisa computer, a forerunner of the Macintosh.

2.4.2.4 Attitude toward using (A)

Attitude toward using, A, is determined by both PU and PEU with relative weights statistically

estimated by linear regression (Davis, 1989; Davis, Bagozzi & Warshaw, 1989). PU has a positive

influence on A, because positively valued outcomes increase the means to achieve the desired

outcomes (Rosenberg, 1956). PEU plays an important role in determining A, because PEU

distinguishes two mechanisms in which PEU influences behavior and attitudes: self-efficiency and

instrumentality (Davis, 1986). The direct relationship PEU-A is present to capture the intrinsically

motivating aspect of PEU (ibid).

2.4.2.5 Behavioral Intent (BI)

Behavioral intent is determined by both the person’s attitude toward using the system (A) and its

perceived usefulness (PU), with weights estimated by regression. The relationship between BI and

PU is based on the idea that people form intentions toward behaviors in which they have a belief

would increase their performance (Davis, Bagozzi & Warshaw, 1989). The desire of the increase in

job performance, are resting in the ability that it will lead to extrinsic rewards, such as pay increase

and promotion (e.g., Vroom, 1964). The relationship between A-BI, implies that “all else being equal,

people form intentions to perform behaviors toward which they have positive affect” (Davis, Bagozzi

& Warshaw, 1989, p. 986).

2.4.2.6 Actual Use

Actual system use is the last parameter of TAM and it has a direct-relationship with BI. Hence, the

actual usage is dependent on our behavioral intent.

25

2.4.3 Theoretical Foundations of Technological Acceptance Model

2.4.3.1 Self-Efficacy Theory

PEU has a strong linkage to the self-efficacy theory developed by Bandura (1982, p. 122), in which

self-efficiency was defined as “judgements of how well one can execute courses of action required to

deal with prospective situations”. Bandura’s theory on self-efficacy makes an important distinction

between self-efficacy judgement and outcome judgement. Outcome judgement is in concern to which

extent a behavior, that has been successfully executed, is believed to be linked to a valued outcome

(Davis, 1989). These two elements, self-efficacy judgement and outcome judgement, can be related

to PEU respective PU. Following, Bandura (1982, p. 140) argues that “in any given instance, behavior

would be best predicted by considering both self-efficacy and outcome beliefs”.

2.4.3.2 Motivational Theory

Motivational theory can be divided into two main types of motivation, extrinsic and intrinsic.

Extrinsic motivation can be explained as the performance of an activity because it is perceived

important to achieve a valued outcome, such as pay, improved job performance or promotions,

distinct from the activity itself (e.g., Lawler & Porter, 1967; Vroom, 1964). On the other hand,

intrinsic motivation refers to the performance of an activity for no apparent reason other than the

process of performing the activity per se (e.g., Berlyne, 1966; White, 1959).

Connecting motivational theory to TAM, PU can be seen as an example of an extrinsic motivation

whereas an example of intrinsic motivation is PE. This can be explained further, an extrinsic

motivated user is driven by the expectation of some benefits or reward external to the system

interaction, whereas the intrinsically motivated user is driven by the benefits from the interaction with

the system per se (Brief & Aldag, 1977). To demonstrate how motivational theory is relevant for

TAM two examples follow. PU is driven by an outside benefit, so to say external to the system

interaction, for example improving job performance, consequently PU is driven by extrinsic

motivation. PE, on the other hand, specifies to what extent joy may be derived from using the system

itself, therefore PE is driven by intrinsic motivation.

2.4.3.3 Cost-benefit paradigm, behavioral decision theory

The Cost-benefit paradigm is part of the behavioral decision theory (Johnson and Payne, 1985; Payne,

1982). This paradigm is relevant to TAM and more specifically PU and PEU because it explains

people’s choice among different decision making strategies. There is a cognitive trade-off between

the effort required to utilize the strategy and the quality, or so called accuracy, of the resulting decision

26

(Davis, 1989). This, the distinction between subjective decision making performance and effort, can

be compared to the distinction between PU and PEU (ibid).

27

3. Methodology

________________________________________________________________________________

This chapter will present a comprehensive overview of how the study has been conducted. First, the

research strategy will be presented, followed by the research design and thereafter the method for

data collection and data analysis is explained. All of these sections will provide the reasoning behind

the choice and the advantages and disadvantages that are in association to the chosen research.

________________________________________________________________________________

3.1 Research Strategy

The research strategy in this thesis consists of a deductive approach, the main philosophical

assumptions objectivism and positivism and the methodological strategy chosen in this thesis is of

quantitative nature.

3.1.1 Deductive Approach

The chosen approach for this thesis is a deductive approach. The approach is characterized by deriving

hypotheses from theory, collecting data, present findings and later confirm or reject the hypothesis

made. At last, some revision of the theory may be present. This is the most common approach when

conducting quantitative research (Bell, Bryman & Harley, 2019, p. 23). Displayed in figure 3.1 please

see the model of deductive approach.

Figure 3.1 The Process of Deduction (Bell, Bryman & Harley, 2019, p. 21).

The research for this thesis was started by mapping out relevant literature by the overall research

topic, hence finding relevant theories within the research topic. After the theories are established,

hypotheses were drawn from the present theories in an attempt to explain a limited aspect of the

business organizations, in specific how decisions best are determined in business organizations. Later

the hypotheses were translated into operational terms in order to explain how data may be collected

in relation to the concepts that make up the hypothesis. Before the general data collection started a

pilot survey was made in order to establish that the questionnaire was of satisfactory level. The data

was collected by students at School of Business Economics and Law took part of the study and later

the findings were mapped. These findings made it possible to reject or confirm the hypotheses. At

last, revision of theory was made in line with the findings in the collected data. This is because that

28

the data found may not fit with the original hypotheses or that the relevance of the dataset may become

visible first after the data have been collected (ibid, p. 21).

3.1.2 Assumptions, Objectivism & Positivism

It is important to be aware of philosophy assumptions, such as ontological or epistemological

considerations when conducting research (ibid, p. 26). The lack of awareness of these assumptions

may lead to that less valuable knowledge is generated by the findings because of the lack of

consistency between the different assumptions (ibid, p. 26).

Bell, Bryman and Harley (2019, p. 26) explains that ontological considerations “is about the

assumptions we make about what it means for something to exist”. Consequently, understanding

ontological considerations such as: objectivism or constructivism is important in order to understand

reality. When there is awareness of our own ontological assumptions, it is possible to design a

research in which most effectively may capture the reality in which we seek to understand (ibid, p.

26). Objectivism can be described as that they have an objective reality, independent of our role as

an observer whereas constructionism is in regards to socially-constructed entities, so to say are made

by the actions and understanding of humans (ibid, pp. 26-27).

The second philosophical assumption that a researcher needs to be aware of is epistemological

considerations. Epistemological considerations can be understood as a particular understanding of

how researchers may gain knowledge of reality (ibid, p. 29). Hence, epistemology allows us to answer

how researchers should conduct research. Epistemological considerations can be divided in two

different main views; positivism and interpretivism. Positivism argues that “because reality exists

objectively and externally, the appropriate way to gather data is to observe phenomena directly or to

‘measure’ them using surveys or other instruments” (ibid, p. 30). Interpretivism is fundamentally

different from positivism and it requires the researcher to graph the subjective meaning of social

action (ibid, p. 31).

This thesis is based upon the ontological consideration of objectivism and the epistemological

consideration of positivism. Consequently, the understanding of what the reality is and how we can

gain knowledge of that reality is set. By the awareness of philosophical assumptions of reality it is

and what method is most suitable, it is possible to mitigate the risk of soundness of the data produced

and the knowledge generated (ibid, p. 33).

29

3.1.3 Quantitative Approach

There are two different methodological research strategies; quantitative approach and qualitative

approach. This research has applied the quantitative approach which is in line with the underlying

philosophical assumptions that are present, so to say objectivism and positivism. Quantitative

research is characterized by the quantification in the collection and analysis of data and it entails a

deductive approach, which emphasizes testing of theories (ibid, p. 35). Figure 3.2 outlines the main

steps in quantitative approach, and the research done in this thesis follows this process.

Figure 3.2 The Process of Quantitative Research (Bell, Bryman & Harley, 2019, p. 21).

Just as most research (ibid, p. 164), the process in which this thesis was conducted did not follow a

strictly linear approach, as figure 3.1 might imply. Thus, this aim of this figure is to capture the main

steps of the research conducted.

The substance in quantitative research is about the ability to measure. Hence, it is important that

concepts derived from different theories are measurable. In order to measure these concepts,

indicators can be classified as either independent or dependent variables. Measure is essential in

quantitative research because it’s underlying philosophical assumptions, earlier established as being

that an objective reality exist and that in order to develop knowledge about reality, there is a need for

being able to engage objectively with the research topic currently examined. One benefit that Bell,

Bryman and Harley (2019, p. 168) highlights with quantitative research is that being able to measure

a concept enables fine differences between subjects to be established. In contrast to qualitative

research, whereas it is often possible to detect between subjects in extreme categories, the fine

differences are harder to detect. Another aspect that measurement enables is the ability to be

consistent over time and being consistent in relation to other researchers (ibid, p. 168), thus changes

in the phenomenon measured is the only real change in the model applied. All concepts may not be

possible to measure directly, hence in this thesis I will refer to the aspects in which the concept can

be derived as indicators.

30

3.2 Research Design

The research design of this thesis will be discussed in the section below. First, by a discussion about

the quality criteria of the research, namely reliability, replicability and validity. Lastly, a review of

the chosen research design, namely cross-sectional design will be argued for.

3.2.1 Quality Criteria

According to Bell, Bryman and Harley (2019, p. 46) three common measurements for evaluating

business and management research in the quantitative research approach are reliability, replicability

and validity. Providing information about the quality of the research is important in order for the

findings to be strengthened and contribution towards knowledge may be presented. Consequently, all

of these aspects will be discussed below in regards to this thesis.

3.2.1.1 Reliability

The quality criteria, reliability, is concerned about whether the results of a study are repeatable or not

(ibid, p. 46). Reliability is important because it refers to the consistency of measuring a concept.

Reliability can be divided into three subcategories, stability, internal reliability and inter-rater

reliability (ibid, p. 172). These three types of reliability will be clarified below and how these have

been accounted for in the thesis.

Stability

Stability refers to whether a measure is stable over time or not. The most common way of displaying

this measurement is to perform a test-retest method (ibid, p. 172). There are however problems related

to this way of testing. How respondents reply at time period one, may influence how the respondents

reply at time period two. Another problem with this method is that events might happen between the

different time periods resulting in problems with consistency may arise (ibid, p. 172). Partly because

of these indicators, but also because the value of anonymity to the respondents a test-retest method

has not been performed in this thesis. Consequently, this part of reliability has not been established.

Internal reliability

Internal reliability is applied when multiple-indicator measures are present, such as in this thesis.

Internal reliability measures “whether respondents’ scores on any one indicator tend to be related to

their scores on the other indicators associated with that concept” (ibid, p. 172). The internal reliability

has been examined in this thesis through the application of Cronbach’s alpha. In a reliable scale, all

items should correlate well with the total (Field, 2018, p. 826), the correlation between all variables

31

in each model had to exceed .3, all data have an item-total correlation above .3, which is encouraging.

The accepted tolerance level was set to .80 (Davis, 1985), consequently the same tolerance level was

applied in this survey data. The internal reliability of this research showed a high quality, three of the

factors (PEU, PE & A) had ∝´s above .80 whereas two factors (PU & BI) exceeded this and showed

an ∝ above .90. Hence, the internal reliability in this study is proven high.

Inter-rater reliability

This reliability measure is most relevant when there is a great deal of subjective judgement involved

such as translation of data into categories there is a possibility of lack of consistency in these decisions

made (Bell, Bryman & Harley, 2019, p. 172). In this thesis, business decision support has been

examined through the application of the framework TAM. Resulting in that the framework is set, and

no decision has been made of how to classify different subjects because the model of the data

collection and how to analyze the result is already set by another researcher.

However, the rapid development in the field of Artificial Intelligence is changing the settings for

decision support in organizations and that might have an effect on the reliability of the study.

Consequently, if a similar study is carried out in the future, there is a risk that the conditions may be

different from today's setting. However, this study was carried out as an initial study to examine

students preferences and their poses to decision support in organization, this study may still be

considered reliable because future research may add to this research and extend on its findings.

3.2.1.2 Replicability

One aspect that is important in relation to the quality criteria is that the study is replicable, so to say,

the study’s findings need to be replicable for other researchers (ibid, p. 46). To ensure the replicability

of this study, the researchers were transparent and careful in the description of how this research was

conducted. Especially section 3.3 is written in great detail to enable the study to be replicated with

similar results in the future. Replicability of the study is important due to checking that the researchers

biases and values not contaminated the study (ibid, p. 178). If an attempt to replication not is

successful it may possibly undermine the validity of the findings in the research that was attempted

to replicate.

3.2.1.3 Validity

Validity is concerned with the integrity of conclusions made in the research (ibid, p. 46). According

to Bell, Bryman and Harley (2019, p. 46) there are four different main types of validity measurement-

, internal-, external- and ecological validity. Out of these four categories external-, internal- and

32

measurement validity was judged as applicable of this study. A discussion of these sub-categories in

relation to this thesis will follow below.

Measurement Validity

Measurement validity is concerned with if the measures really represent the concepts that they are

intended to represent (ibid, p. 46). Measurement validity can be distinguished even further, by

dividing it into sub-categorizes, face validity, concurrent validity, predictive validity, convergent

validity and discriminant validity. The one of these sub-categorized that is used in this research is

predictive validity. Predictive validity can be described as asking respondents whether they are likely

to perform a certain activity in the future (ibid, p. 175). This has been done in the questionnaire in

this thesis, when collecting the respondents intended use of decision support system in their job

position in the future, please see appendix A.

Internal Validity

Internal Validity raises the problems with causality, whether we can be sure or not that the causal

relationship is genuine and not produced by something else (ibid, p. 47). Causality will be referred to

in this thesis as the factor that has a causal impact, independent variable and the effect as the

dependent variable. The framework for this thesis is TAM, and some causal relationships have been

set between the variables. These have not been determined by the researcher performing this thesis,

but with a tested validated model and hence will be used without any precautions about the variables

causality. Hence, this validity measured has been sought after and it is not presumed to be of any

concern for this thesis.

External Validity

This is one of the most relevant validity measures of this thesis. External validity is to concern whether

the findings of the study can be generalized beyond the specific topic in mind (ibid, p. 47). As a

quantitative researcher, it is crucial that the sample is representable for the entire population in order

to validate the quality of the findings. The external validity is considered strong with the sample from

which the data collected has been randomly selected (ibid, p. 59). In the conducted study, the sampling

was not random, more on this later on section 3.3.2.2 sampling. Thus this raised some concern in

relation to the overall external validity of this thesis.

3.2.2 Cross-Sectional Research Design

Cross-sectional research design is characterized by that the answers on the different variables are

given at a single point in time and more than one case, most often employ quantitative or quantifiable

33

data and the only relationships possible to examine is between the variables (Bell, Bryman & Harley,

2019, p. 58). This design was chosen because it entails collection of data at a single point in time

referring to the fact that the data is collected more or less simultaneously. The approach of this

research was to examine students' behavioral intent of using AI tools to make decisions in the future,

so to say, their perception at a single point in time. Consequently this method was perceived as

suitable. The aspect of “cases” refers to people, or students, because their perception is crucial in

order to understand how AI might be used in decision making in organizations in the future. Derived

from the different “cases” data was collected and examined in order to find relationships and patterns

concerning these individuals.

The study carried out was of quantitative approach which is suitable with cross-sectional design

because this design enables collection of large amounts of data, hence this design was considered

appropriate for this research. Please note that the aim of establishing variation is only possible when

several cases are used, however establishing causality of the variables have proven difficult. More on

causality, please see section “internal validity” above whereas it is discussed in regards to this thesis.

3.3. Research Method

3.3.1 Secondary data collection

The secondary data collection was initially started with the conducting of a systematic literature

review. According to Bell, Bryman and Harley (2019, pp. 92-96) conducting a systematic literature

review is preferable because the actions, such as decisions, procedures and conclusions, taken by the

researcher can be traced and biases may be limited. The systematic literature review was based on the

formulated research question, that the quality of the found research was of satisfactory level and the

information found was sufficient in order to answer the defined research question.

A secondary data collection was conducted in order to acquire information on existing research on

the different fields of interest in order to understand and be able to answer the research question.

According to Bell, Bryman and Harley (2019, p. 91) benefits of using secondary data is that it is an

easily accessible and time-efficient source of high quality data. To start the secondary data collection

off, literature related to the technology acceptance model was collected. This because it felt like a

necessity to understand the theoretical model of which the primary data collection questionnaire

would be based upon. This area is fundamental for both conducting and to answer the research

question. Later the literature of decision making theory and Artificial Intelligence was collected in an

attempt to map the intersection between those areas and understand whereas the field is currently. At

34

last, literature related to decision support system, the intersection of focus between AI and decision

making, and drivers of technological advancements was mapped in order to connect the theoretical

findings.

The mixture between the different sources used throughout the theoretical framework is mainly

articles and books with an emphasis on the first. However, to complement the printed material

collected, dictionaries, government papers, consultancy reports and newspapers published online

were used as a complementary resource. The two following subcategories below will describe which

databases and keywords that were used in the systematic literature review.

3.3.1.1 Databases

Two main databases were used when collecting secondary data; Scopus and Business Source Premier.

These two were used because the author was familiar with the databases and the databases have a lot

of articles in the field of business and economics. Scopus, which was the one most used in the

collection of secondary data, only provides peer-reviewed literature and it is user friendly because the

results can be based on data or relevance. Being able to sort the findings depending on the date of

publication is important in a field that is rapidly undergoing changes, such as AI in order to receive

the latest information available on the field. In order to download the articles, when there was no

direct link connected from the database, Google scholar and Gothenburg University Library.

3.3.1.2 Key words

The following keywords were used in different combinations: Technological advancements,

Technology, Big data, Data driven decision making, Decision Making (or DM), Decision Support

System (or DSS), Decision support, Decision aid, Intelligent decision support system, Artificial

Intelligence (or AI), Automation, Risk, Technology Acceptance Model (or TAM), History,

Development, Definition, Concepts and Organization (or Organisation).

3.3.2 Primary data collection

The primary data was collected through a quantitative study, this approach was chosen because it can

be an useful approach when a large body of empirical data has to be collected. This method of

collection was appropriate because the researcher wanted to gather a large number of respondents.

The quantitative study was based upon a self-completion questionnaire.

The questionnaire was based on Davis “technology acceptance model” and aimed at investigating

students’ perception of their behavioral intent of using AI tools in relation to decision making at their

35

future workplace. The behavioral intent refers to the different components described in section 2.4.2,

such as PEU and PU but these PE, anticipated use of DSS and perceived characteristics of output.

According to Bell, Bryman and Harley (2019, pp. 233-235) there are different aspects to consider

when conducting self-completion questionnaires. This approach is convenient for the respondent

because they can choose wherever and whenever they want to complete it. The “interviewer effect”

is eliminated because there is no interviewer present. Consequently, the risk of having answers be

affected by the interviewer or the interviewee answering according to what he or she believes the

interviewer wants to hear. Another aspect that is eliminated because there is no interviewer present,

is the risk of “interviewer variability”. Self-completion questionnaire does not suffer due to the

interviewer asking questions in different orders or in different ways that might be a present risk in

other collection methods.

However, there are also negative aspects that it is important to be aware of when approaching a self-

completion questionnaire (ibid, pp. 233-235). First, there is no one present to answer any questions

that might arise during the process. Second, the problem of missing data is present in a self-

completion questionnaire. Third, there might be a risk of respondents finding the topic salient and

thus time consuming or dull. Hence, it is of great importance to produce well thought after questions

so that minimizes the risk present that the respondent chooses to leave the survey without sending the

result in, or answer the survey in a careless way. Fourth, there is a risk present in regard to the response

rate. However, during the time period this study was conducted, the virus Covid-19 was present and

distributing the questionnaires in person to class after class was not an option even if it could have

possibly increased the response rate.

3.3.2.1 Questionnaire Design

As earlier mentioned, the questionnaire was based upon research conducted by Davis and the

framework TAM was applied to an AI tool, namely DSS. In an attempt to mitigate the possible risks

with the method of choice, the design of the questionnaire was crucial. It was important that the

questions were clear and unambiguous. In order to address this matter, the self-completion

questionnaire was sent out to review and to alternate before the final version was released. It was

important that the questionnaire was easy to follow, and easy to understand and answer.

The questions were closed and the majority of the questions had a horizontal format. Closed questions

were used because of several reasons, it is both convenient for the respondents that did not have to

write long answers, but also less effort is necessary on coding which is a benefit for the researcher

36

(ibid, pp. 237-239). In order to reduce the risk of fatigue of the respondents, the questionnaire should

be as short as possible. However, this questionnaire was based upon a framework the researcher was

limited in how much could be excluded without losing value of the study. Despite the fact the number

of questions were set, the choice of having horizontal format in the questionnaire played a role,

because the questionnaire is perceived shorter in length (ibid, pp. 237-239). In the questionnaire

design it was made sure that no questions could be left blank in order to reduce the risk of missing

data for the variables. In order to increase the respondents willingness to answer the survey, the

answers were anonymous.

3.3.2.2 Sampling

The approach of the sampling was non-probability sampling, in more detail convenience sampling.

The population was defined as students undertaking a bachelor’s or master’s programme at School of

Business Economics and Law, University of Gothenburg. The size of the population was received

from Ladok, which is a platform in which all students need to be registered, and the size of the

population was 3655 students. Furthermore, the questionnaire was sent out to approximately 600

students’ and the contacts were provided by the school. Hence convenience sampling was used

because the school send out an invitation the questionnaire to a mail list in which they already had

constructed. Convenience sampling refers to that it is simply available to the researcher (ibid, p. 197),

the school that provided and sent out the questionnaire, thus this may be an example of convenience

sampling being used. Consequently, the sample size was 600 thus meaning that 1 out of 16 students

received the questionnaire. Bell, Bryman and Harley (2019, p. 197) argues that when the population

is homogeneous, such as in a company or an occupation, the amount of variation is less leading to

that the relevance of a larger sample group is less important. All participants of the study are students,

hence the homogeneity is large in the population which makes the size of the sample size less crucial.

A screening process was conducted in the self-completion questionnaire in order to make sure that

only people eligible to the study took part of it, see appendix A. In total there were 82 answers,

whereas one person no longer was a student at School of Business Economics and Law, University

of Gothenburg. Consequently, this response will not be part of the data collection it has been sifted

out of the study. The response rate of the questionnaire was 13,5% which is perceived as quite low.

However, because the nature of the sample, only contacts provided by school and that they did not

agree on taking part of the study in beforehand, that the response rate would be quite low was not

unexpected but the researcher did not expect the response rate to be this low.

37

From the data received from Ladok, 3655 students are currently undertaking a bachelor’s or master’s

programme, out of these were 45,6% males and 54,4% were females. This can be compared to the

answers to my questionnaire, in which 44,4% were males and 55,6%. None of the respondents

answered “prefer not to say” or “other” which were available in the questionnaire.

More than half of the respondents answered that they are between 24-26 years old (46), the second

largest age group were 27-30 years old (13) followed by 21-23 years old (12). The age group with

the fewest respondents were 18-20 years old which only had 2 respondents, which is in line with the

fact that you normally do not start University until you are 19 years old at earliest, with a few

exceptions. Consequently, these two individuals are most likely in their first or second year of studies

in the University.

The biggest group in terms of completed education was not too surprisingly bachelor’s degree with

65,43%. The relatively high number of respondents answering master’s degree 24,69% is a bit

worrying. Maybe there are students in their second year of studies at master’s level that perceive

themselves as more or less done with the master degree or maybe the respondents misinterpreted the

question and assigned the level of studies they are currently undertaking to the questionnaire instead

of the level they actually have completed. That only eight respondents have answered high school,

follows the same pattern, maybe people still in the bachelor answered that they were pursuing a

bachelor’s degree instead of that they have completed the high school diploma. None of the

respondents answered “PhD or higher” which is a good indication that the group of students sought

for were reached.

In the aspect of gender, the responses were quite well in relation to the population. However, age or

the level of completed education will not be compared to the population because the population data

only concerns gender, because that was the statistics Ladok, could provide. However, because the

way of sampling is done, a non-probability sampling method, it can be questioned how generalizable

the findings are because of the problem with representative samples. But the only aspect that could

be compared to the population, gender, proved to be representing the population to a satisfactory

level.

3.4 Data Analysis

Initially a reliability test was conducted in order to check the internal reliability of the measures used

in the model. The minimum reliability set out to achieve was .80, as Davis (1985, p. 96) set in the

original model.

38

In the second step a factor analysis was conducted. The aim was to construct one factor each out of

the groups of variables. However, the factor analysis model was constructed by Eigenvalues and not

by fixing the number of factors in the model, this because the researcher wanted to examine whether

the variables could be grouped into several factors or to see if the model holds and only one factor is

extracted from each group of variables. The eigenvalues were set to only retain factors with

eigenvalues greater than one, the so called Kaiser’ criterion (Field, 2018, p. 790). The model was set

to direct oblimin in SPSS statistics, referring to that correlation between the variables is allowed,

however because only one factor was constructed from each group of variables the solution cannot

be rotated. Because of the correlation between the variables, and the model set thereafter, any

correlation between any variables less than .3 is to be excluded (ibid, p. 798). By performing the

factor analysis, loading values for each factor is generated enabling the researcher to establish the

relationship between the components of the technology acceptance model. KMO (or Kaiser-Meyer-

Olkin) is a measure of sampling adequacy and a value closer to one that refers to the fact that the

factor analysis should yield a distinct and reliable factor (ibid, p. 798). In this model, a KMO value

above 0.5 is set to be acceptable. Beta values with an absolute value greater than .4 will be interpreted

as suggested in Field (2018, p. 795).

Later a regression was carried out in order to detect the point estimates of the factors relationships

and to test the hypotheses made in introduction. The confidence interval level in the regression was

set to 95% corresponding to Davis (1985). To measure the accuracy of the model, if the test was to

be carried out in the entire population instead of the sample (Field, 2018, p. 411), the relationship

between R2 and adjusted R2 was determined in the different regression models constructed. The

standardized beta was used in the regression analysis, because contrary to beta, they are not dependent

on the units of measurement of the variable (ibid, p. 415), hence easier to interpret.

The Technology Acceptance Model framework and its building block were used throughout the

process in order to analyze and contrast the theoretical and empirical findings. The empirical findings

were matched with theoretical findings and derived from this, the conclusions emerged.

39

4. Empirical Findings

________________________________________________________________________________

This chapter will present the result of the study conducted. First, a brief overview of the internal

reliability of the model is presented using Cronbach Alpha as the method of testing the reliability.

Second, a comprehensive view of the factor analysis conducted is explained. At last, a presentation

of the result of the regression carried out is presented and estimates of parameters are provided.

________________________________________________________________________________

4.1 Internal Reliability Test, using Cronbach Alpha

As explained in section 3.4, Cronbach Alpha was used in order to test the internal reliability of the

model. The achieved reliability surpassed the .80 that was set in all of the variables, and also exceeded

.90 in perceived usefulness (PU) and behavioral intent (BI), see table 4.1. As discussed in the section

about the internal reliability, the correlation between the variables had to exceed .3, all data have an

item-total correlation above .3, which is encouraging. All in all, this indicates good reliability in the

model.

There were in total four questions in the questionnaire about behavioral intent, see appendix A.

However, only variables one and two will be used because they are based on the scale 1-7 as the other

variables. The other variables, needed to be recoded into 1-7 scale items in order for them to be

applied in the factor analysis. Because, of the complexity of performing this action, the first two

variables for BI will be used, and variable three and four will not be included in the model. Hence,

the item column in table 4.1 below will only show 2 items for behavioral intent.

Item two and four in attitude towards using (A) was recoded because it had harmful and negative to

the left hand side of the likert scale and beneficial and positive on the right hand side. Because all

positive words were put on the left hand side of the likert scale, these needed to be recoded in order

to not generate a negative alpha.

40

Variable Label #Items

Cronbach Alpha

Reliability

Perceived Ease of Use PEU 4 .844

Perceived Usefulness PU 4 .919

Perceived Enjoyment PE 3 .820

Attitude Toward

Using A 4 .825

Behavioral Intent BI 2

.949

Table 4.1 Cronbach Alpha Reliability of Measurement Scales

4.2 Factor Analysis

In order to establish how the different factors and their variables affect behavioral intentions among

students towards using DSS as a decision making tool a factor analysis was conducted. In example,

PU consists of four different questions, so to say variables and together they form the factor PU.

A factor analysis was run in SPSS statistics and the following requirements were set to the model.

Initially, the correlation between the different group of variables were sought after in order to ensure

that the choices in constructing the model is accurate. Furthermore, if correlation between any

variables were less than .3 one of the variables causing this is to be taken away from the model,

however this was not the case, see Appendix B.

In order to extract a factor, the eigenvalue had to be greater than one, the rotation of the model was

set to direct oblimin, meaning that correlation between the variables is allowed, and the variables all

correlated to a satisfactory level. However no rotation was done because only one factor per each

group of variables was extracted. At last, KMO was acceptable at a level of .5 which all factors proved

a value above. Table 4.2 indicates the loading of the relationships established between the variables

and the factors. As explained in section 4.3, loading values exceeding .4 in absolute value will be

interpreted.

41

Construct Mean StD Indicator Loading, β

PEU1 2.8395 1.29862 .616

PEU2 3.5432 1.21500 .778

PEU3 2.9383 1.25843 .875

PEU4 3.3210 1.13828 .781

PU1 2.6790 1.28284 .849

PU2 2.5432 1.29469 .847

PU3 2.5556 1.14018 .868

PU4 2.4938 1.23616 .883

A1 2.4568 1.19425 .682

A2 2.3704 1.21906 .838

A3 2.4691 1.11900 .911

A4 2.6296 1.35503 .563

BI1 2.7037 1.31762 .951

BI2 2.6296 1.25941 .951

Table 4.2 Measurement Statistics of Variables

The measurement statistics of the factors created in the factor analysis are displayed below in table

4.3. The mean refers to the seven likert scale in the questionnaire whereas the lower the mean,

indicates that the students’ perceive the factor more likely to influence whether they will use the DSS

or not. The mean refer to the number one to seven at the likert scale in the questionnaire, whereas

they refer to: extremely likely (1), quite likely (2), slightly likely (3), neither (4), slightly unlikely (5),

quite unlikely (6) and extremely unlikely (7). Table 4.3 show that the mean varies from 1.85 to 2.54

referring to that a positive perception of students willing to use DSS as a decision making tool exist

in the data set.

42

Construct Mean StD

PEU 2.4176 .77755

PU 2.2122 .95824

A 1.8479 .74622

BI 2.5360 1.19586

Table 4.3 Measurement Statistics of Factors Constructed from Variables

4.3 Regression Test & Parameter Estimates

Table 4.4 provides an overview of the OLS regressions carried out in order to confirm or reject

hypotheses. Which independent variables that will be included for each dependent variable is given

from the constructs of TAM framework, see figure 2.2. As explained in section 3.4, the difference

between R2 and adjusted R2 is aiming to be as small as possible. In example, if the difference is (.170-

.160) it accounts for approximately 1% less variance in the outcome if the model were to be derived

from the population rather than the sample. The model would have approximately 1% - 1,6% less

variance in the outcome if it were to be derived from the population rather than a sample.

Beta, β, may be interpreted as if PEU increases by one standard deviation (.126), see table 4.5, PU

increases by .412 standard deviations. The significance level accepted in this model is p < .05, because

of the framework. The significance level is the probability of rejecting the hypothesis when it is true,

this level of significance indicates that there is a 5% risk of concluding that a difference exists when

there is no actual difference.

43

Dep. Var. R2 Adj. R2 Ind. Var. β t. stat Sig. Lvl.

PU .170 .160 Constant 3.067 .003

PEU .412 4.024 .000

A .495 .482 Constant 3.085 .003

PEU -.002 -.025 .980

PU .704 7.971 .000

BI .379 .363 Constant 2.227 .029

PU .428 3.411 .001

A .234 1.861 .067

Table 4.4 TAM Regression Tests

Table 4.4 establishes that four of the hypotheses may be accepted (H1, H2, H4 & H5) and one rejected

(H3). The last hypothesis (H6) will be explained after table 4.6 is presented. Hypothesis one (PEU

will have a significant effect on PU) was supported. Table 4.4 displays that PEU has a significant

effect on PU with the level of p < .001. Meaning, that if the student finds the use of the decision

support system easy to work as an aid in decision making, the student is more likely to find it useful

as well. β, may be interpreted as if PEU increases by one standard deviation (.126), see table 4.5, PU

increases by .412 standard deviations. The second hypothesis (PU will have a significant effect on A,

controlling for PEU) was also supported. This relationship was found significant at a level of p <

.001, see table 4.4. This indicates that if the student finds the decision support system to be useful to

work as an aid in making decisions, the student will develop a favorable attitude toward it. β, may be

interpreted as if PU increases by one standard deviation (.103), see table 4.5, A increases by .704

standard deviations. Hypothesis five (PU will have a significant effect on BI, controlling for A) was

also supported. This relationship was found significant at a level of p < .001, see table 4.4. This

indicates that if a student that uses the decision aid perceived it as useful, the student will have a

positive intention toward using the system in decision making. β, may be interpreted as if PU

increases by one standard deviation (.113), see table 4.5, BI increases by .428 standard deviations,

accounted for A.

Hypothesis three (PEU will have a significant effect on A, controlling for PU) was rejected. Table

4.4 indicates that this relationship is significant at a level of p < .980. The level of significance

acceptance was set to, p < .05, the effect PEU has on A, accounted for PU, is consequently not

significant counter to what the hypothesis were made out.

44

Hypotheses four want to test if some of the relationships would not have a significant effect on

another. The fourth hypothesis (A will not have a significant effect on BI, controlling for PU) was

also supported. Table 4.4 indicates that this relationship is significant at a level of p < .067. The level

of significance acceptance was set to, p < .05, the effect A has on BI, accounted for PU, is

consequently not significant.

Table 4.5 provides the point estimates and the confidence intervals based on regressions models in

which only two variables were put into the model, the independent and the dependent variable. The

causal perspective highlights the fact that PEU may be an antecedent to PU, rather than a parallel. In

order to assess which regression model that is preferred, a causal regression is carried out to be

compared to the above regression when several independent variables were controlled for.

Causal Link Point Estimates

Ind.

Var.

Dep.

Var.

Adj.

R2 β

StD.

Error

Sig.

Level

PEU PU .160 .412 .126 .000

PEU A .071 .288 .103 .009

PU A .488 .703 .062 .000

PU BI .343 .593 .113 .000

A BI .277 .535 .152 .000

Table 4.5 TAM Causal Regression Test

Table 4.6 highlights the different adjusted R2 values in the different regression models constructed,

both when including several independent variables but also the variables separately. The information

in table 4.6 is derived from table 4.4 and table 4.5. The relationship between PEU and A was not

found significant when controlling for PU. The adjusted R2 when the causal relationship between A

and PEU is examined proved to be very low, hence PEU explains the variation in A poorly. However,

the model when both variables, PU and PEU, were included was not the highest in adjusted R2, hence

PU gives the best model in determining A. This indicates that PEU can be seen as a predecessor of

45

PU and not a determine in directly influencing A. A was not found significant in affecting BI when

controlling for PU, and when R2 is compared in the models below, A explains the variation in BI less

than PU or the when accounting for both variables in the model. However, in this case the best model

is when both PU and A are accounted for.

Dep. Var. Ind. Var. Adjusted R2

A PEU .071

PU .488

Both are included .482

BI PU .343

A .277

Both are included .363

Table 4.6 Compilation of Adjusted R2 in Different Regression Models

Hypothesis six (PU will have a greater impact on BI than PEU) was confirmed. The direct relationship

between PEU and A showed that PEU is not a good determinant of the explanation is A. Furthermore,

A (which is a component of PU and PEU) did not explain the variation in BI well, but PU showed

significant in hypothesis two and five and also showed to be a good predictor of both A and BI in the

direct relationship. Furthermore, figure 4.1 enables calculations of the relative importance of the

different factors in influencing behavioral intent. PU has both a direct (.593) and an indirect effect

via A (.703.535) in influencing BI, combined this refers to .969. Whereas PEU has an effect through

PU (.412*.969) resulting in an effect of .399. Comparatively, PU is therefore 2,43 times as important

in influencing behavioral intent in the survey data. Hence, PU have a greater impact on BI than PEU.

4.1.2 Causal diagram of Technology Acceptance Model

The findings above indicate that PEU is a predecessor of PU and the constructs of TAM would look

like this if the beta values of the regressions are inserted into the model. This figure aims to provide

an easy overview of the relationships in the survey data. This will enable other researchers to compare

this figure to their own findings about the constructs of TAM.

46

Figure 4.1 Diagram of Technology Acceptance Model

47

5. Discussion

________________________________________________________________________________

This chapter starts by applying the TAM into the gathered data in order to structure and analyze the

findings. The researcher first explain the main findings, and later displays the findings in relation to

the different components of TAM. Simultaneously, the findings are set in relation to the theoretical

framework in order to deepen the level of discussion.

________________________________________________________________________________

5.1 Main Findings

The internal reliability test performed of the data set indicated that there were a good internal

reliability in the model, see table 4.1. Whereas, PU and BI where the components which the highest

result of the Cronbach Alpha test.

The factor analysis showed that each group of variables (PU, PEU, A, BI) each could be categorized

into one factor with different loading indicators, see table 4.2. The fact that they naturally formed one

factor each among the groups, even if the model of the factor analysis allowed more than one factor

to be formed ensured the quality of the variables and the correlation between them as strong. The

measurement statistics of the factors created from the variables is indicated in table 4.3. The mean

refer to the number one to seven at the likert scale in the questionnaire, whereas they refer to:

extremely likely (1), quite likely (2), slightly likely (3), neither (4), slightly unlikely (5), quite unlikely

(6) and extremely unlikely (7). Table 4.3 show that the mean varies from 1.85 to 2.54 referring to that

a positive perception of students willing to use DSS as a decision making tool exist in the data set.

Regressions were carried out in order to test the hypotheses made in the introduction, same as Davis

concluded true in the model back in 1985. These regressions were carried out with more than one

independent variable (in exception to H1), because the research wanted to examine whether the

variable had a significant effect of a certain variable controlling for another variable. The test

confirmed first four of the hypotheses, (H1: PEU will have a significant effect on PU), (H2: PU will

have a significant effect on A, controlling for PEU), (H5: PU will have a significant effect on BI,

controlling for A) and (H4: A will not have a significant effect on BI, controlling for PU), see table

4.4. Hypothesis three, (PEU will have a significant effect on A, controlling for PU) was rejected. The

result of this are that PEU and A does not have a significant effect of PU respectively BI controlling

for PU.

48

The relationship between the different factors were also tested when not controlling for another factor,

see table 4.5. The result of this showed that all variables were significant to the model at the desired

significance level, when not controlling for another variable. However, in order to discuss the

relevance of this test the R2 value in the different models were compared. Table 4.6 displays that R2

was the highest when only PU was seen as a determinant of A and when both PU and A where seen

as a determinant of BI. Hence, the construction of figure 4.1 where PEU is seen as a predecessor of

PU rather than a parallel and PEU are displayed as not having an direct relationship with A, as first

indicated in figure 2.2.

The point estimates derived from the regressions, enabled calculation of the relative importance of

different factors in influencing behavioral intention. This enabled calculation of the relative

importance of PU and PEU in affecting BI. PU was 2,43 times as important in influencing BI than

PEU in the survey data, hence hypothesis six was also confirmed (H6: PU will have a greater impact

on BI than perceived ease of use PEU).

The construct of the causal diagram of TAM enables an accessible view of the relationships

established in table 4.2, 4.4. and 4.5. As Davis concluded in studies conducted in 1985 and 1989,

there are two main beliefs that explain user acceptance, PU and PEU. PU is related to “the degree to

which a person believes that using a particular system would enhance his or her job performance”

(Davis, 1989, p. 320) and PEU as “the degree to which a person believes that using a particular system

would be free of effort (ibid, p. 320). As the causal diagram, see figure 4.1, suggests is PU and PEU

the main building blocks of the model leading up to attitude toward using (A) and behavioral intention

(BI). A more detailed view of the different components and how they affect the behavioral intention

of students will be displayed in below sections.

5.1.1 Perceived Usefulness

The idea of PU is that if people believe that an application or a technological advancement will help

in order to enhance their job performance they tend to use it. As the model implies, see figure 2.2

alternatively figure 4.1, PU may be affected by both perceived ease of use and external variables. The

factor PU is compounded by four variables (PU1, PU2, PU3, PU4). Whereas PU1 (improve job

performance), PU2 (increase productivity), PU3 (enhance effectiveness) and PU4 (useful), see

appendix A. The factor analysis showed that PU4 had the highest influence of the construct of PU

followed by PU3, PU1 and lastly PU2. The factor PU constructed had a mean value of 2.21 referring

to somewhere between “quite likely” and “slightly likely” on the likert scale, with more emphasize

on the first.

49

Two regressions were carried out with PU as the independent variable. The first was in relation to

H2, whereas the significant effect PU had on A was tested controlling for PEU. This relationship was

found significant at a level of p = < .001. The second was in relation to H5, whereas the significant

effect PU had on BI was tested controlling for A. This relationship was also found significant at a

level of p = < .001. Hence, PU is a strong determinant in this model. The R2 value in the different

regression models were analyzed in order to calculate PU’s and PEU’s influence in behavioral intent.

PU had a combined effect of .969 and PEU .399, resulting in that PU is 2,43 times as important in

influencing behavioral intent in this study.

As mentioned in the theoretical framework, PU can be seen as based upon an extrinsic motivational

basis. Driven by the expectation of some benefit or reward external to the system interaction (Brief

& Aldag, 1977). Consequently, as PU is an important determinant of user acceptance, it is important

to understand how companies may exploit the knowledge of pursuing extrinsic motivations to reward

appreciated behavior. Most companies have experienced having personnel that are unwilling or

reluctant to change the way they work, having the right personnel is crucial (Rosenberg, 2004, p. 5).

The technology on its own does not have a direct value, unless it is accepted and used by the personal

across the company. Hence, having the right personnel is important when rapid technological

advancements requires new systems to be managed.

5.1.2 Perceived Ease of Use

The idea of PEU is that even if the application might be useful, if the system is too hard to use there

is a risk that the performance benefits are outweighed by the effort of using the application (Davis,

Bagozzi & Warshaw, 1989). The easier the application is perceived in relation to another application,

the more likely it is to be accepted by users, all other things equal. However, the result of this study

implies that the importance of PEU is less than first perceived by Davis in 1985. As the model implies,

see figure 2.2 alternatively figure 4.1, PEU may be affected by external variables. The factor PEU is

compounded by four variables in this study (PEU1, PEU2, PEU3, PEU4). Whereas PEU1 (learning

to operate), PU2 (get DSS to do what I want), PU3 (easy to become skillful) and PU4 (find easy to

use), see appendix A. The factor analysis showed that PEU3 had the highest influence of the construct

of PEU followed by PEU4, PEU2 and lastly PEU1. The factor PEU constructed had a mean value of

2.42 referring to somewhere between “quite likely” and “slightly likely” on the likert scale, with more

emphasize on the first, however not as much emphasizes as on PU.

50

The study showed that PEU had a significant effect on PU with p < .001. Meaning that the application

is found more useful if it is perceived as easy to use. Hypothesis three, PEU will have a significant

effect on A when it is controlled for PU was rejected (p < .980). However, table 4.5 displays that PEU

has a significant direct effect on A, when it is not controlled for PU at a level of p < .01. Hence, PEU

is not totally irrelevant in explaining A. However, since R2 is larger when only PU is seen as a

determiner of A, PEU is seen as a predecessor of PU and not directly in influencing A.

As earlier discussed, PEU did not influence behavioral intention of students as much as PU did.

However, in today's business environment a lot of time and money is spent to develop as user friendly

tool as possible. Maybe some of this resources would be more efficient used in explaining the

usefulness of the system, because this have a bigger influence of the behavioral intent, just as it did

in Davis study. However, because that PEU is a building block of PU, it cannot be seen as unimportant

in explaining behavioral intention. In the self-efficacy theory developed by Bandura (1982, p. 140)

he argues that behavioral would be best predicted by considering both “self-efficacy judgements”

(PEU) and “outcome judgements” (PU).

5.1.3 Attitude Toward Using

Attitude toward using is determined by both PU and PEU with relative weights statistically estimated

by linear regression (Davis, 1989; Davis, Bagozzi & Warshaw, 1989). The idea of A is that if people

have a positive attitudes toward using the application, the behavioral intention is positive. As the

model implies, see figure 4.1, A may be affected by both PEU and PU. This by PEUs indirect relation

through PU. The factor A is compounded by four variables (A1, A2, A3, A4). Whereas A1 (using

DSS, good or bad), A2 (using DSS, beneficial or harmful), A3 (using DSS, wise or foolish) and A4

(using DSS, positive or negative), see appendix A. The factor analysis showed that A3 had the highest

influence of the construct of A followed by A2, A1 and lastly A4. The factor A constructed had a

mean value of 1.85 referring to somewhere between “extremely likely” and “quite likely” on the likert

scale, with more emphasize on the second.

Attitude toward using, just as hypothesis four implied, did not have a significant effect of behavioral

intention when it was controlled for PU, p < .01. However, when the causal relationship between A

and BI was established it was found significant at a level of p < .001. The R2 values of the different

models were compared, and the R2 with the highest value, corresponding to when both A and PU was

used as a determinants of BI. Hence, this relationship, between A and BI is kept in figure 4.1.

51

5.1.4 Behavioral Intent

Behavioral intent is determined by both the person’s attitude toward using the system (A) and its

perceived usefulness (PU), with weights estimated by regression, see table 4.4. The relationship

between BI and PU is based on the idea that people form intentions toward behaviors in which they

have a belief would increase their performance (Davis, Bagozzi & Warshaw, 1989). As explain in the

section about PU, the desire of increase in job performance rest in that the belief that will lead to

extrinsic rewards (e.g., Vroom, 1964). The direct relationship between A and BI implies that all else

being equal, people for intentions to perform behaviors toward which they have positive affect (Davis,

Bagozzi & Warshaw, 1989).

The direct relationship between BI and PU were significant at a level of p < 0.001 when controlling

for A, whereas the relationship between BI and A not were found significant to a satisfactory level,

when controlling for PU. However, because the R2 value is bigger when both factors were included

in this model, this is accounted for in figure 4.1. PU was accounted to be 2,43 times as important than

PEU in determining BI. Hence, this is important to take into account when systems are designed.

DSS is a tool that can help people at decision making positions in organizations to make better

decisions. DSS is based on AI and the availability of amounts of data needed to be processed. These

data can be translated into information that is valuable for companies when making decisions (Pereira,

Belo & Ravesteijn, 2018). This has the potential to lower organizational costs and increase the speed

of processing information (Davenport, 2014, p. 115). Thus, enabling new data and models to arise

which will result in better decision making in organizations.

The behavioral intention among students at School of Business Economics and Law, University of

Gothenburg to use DSS in their future workplace is perceived as fairly high. PU is seen as the single

one most important factor in determining the behavioral intent among students perception of using

DSS in their future workplace. The effect PEU has is not unimportant, because it explains PU. To

conclude, the relative importance in influencing BI is larger in PU than PEU.

5.2 Comparison to Davis Original Model

In Davis (1985), original construct of TAM there is one relationship that differ to the findings in this

thesis. This is the relationship between perceived ease of use and attitude toward using. Davis found

that perceived ease of use is significant in affecting attitude toward using when controlling for

perceived usefulness, p < .05. Table 4.4 displays that PEU is not found significant when controlling

for PU in affecting A. Hence, the direct relationship between PEU and A is not included in figure 4.1.

52

It is interesting to discuss why this relationship between PEU and A, when controlling for PU

differentiate from the original findings by Davis. It could be argued that perceived ease of use were

seen as more important before the digitalization era took place. Today people use technological

advancements such as internet every day and thus indicating that user systems are not that hard to use

resulting in that maybe the importance of perceived ease of use no longer holds. However, it is

important to point out that the respondents of the questionnaire of this thesis has not yet been

introduced to a decision support system, the answers in the survey is only based upon their perception.

5.3 The Future Decision Environment in Organizations

Despite the advantages of AI as a decision making tool it is today not used to the expected extent.

This study shows that the behavioral intention among students of using decision support system in

their future workplace is strong. Hence, there is a need of understanding the business environment

and how this might affect it and how changes can be derived. The large amount of data available for

managers, in combination with the rapid technological advancements in AI drives how decisions are

made, less on intuition and more data based (Brynjolfsson & McElheran, 2016). This in combination

of the restrictiveness of the human mind, it can be argued that the old way businesses made decision

are about to change.

As human has proven unreliable as decision makers (e.g., Kahneman et al., 2016; Marwala &

Hurwitz, 2017; Sharda, Delen & Turban, 2018) the level of automated decisions or hybrid decisions

are about to increase. As an organization in this changing business environment it is important to stay

up to date on these rapid changes taking place, and adapt in order not to fail. The result of this study

provides information that the future decision makers of the businesses in the Gothenburg region are

ready to use AI in order to improve decision making in organizations. Hence, now it is up for the

organizations to be ready for switching to more AI as a decision making tool.

However, it is also important to discuss the possible effects of AI systems taking on a bigger role in

decision making. Because it is crucial that humans and AI systems peacefully coexist in

organizations. Burgess (2018, pp. 6-8) argues that AI might have a severe impact of jobs, but that AI

also provides an opportunity that the work will be more effective and the use of their time more

meaningful. Haenlein and Kaplan (2019) argues that regulation is one way to avoid unemployment

at a larger scale cause by the improvements in AI. It is clear that AI is a huge disruptor and will affect

several aspects in our lives (ibid, pp. 6-8). But to what extent automation of decision making in

organizations will take place can only future tell.

53

6. Conclusion

________________________________________________________________________________

This chapter aims to answer the research question. Second, implications in the research conducted

are presented. At last, a suggestion for future research is presented.

________________________________________________________________________

6.1 Main Findings

The purpose of this research were to examine the behavioral intent among future decision makers to

use Artificial Intelligence as a decision making tool by applying the framework Technology

Acceptance Model. The research focuses on students in Sweden, within the field of business and

management and the research question was:

What factors in TAM influence business students’ behavioral intent

to use Artificial Intelligence in decision making?

This study established that perceived usefulness (PU) is 2,43 times as important than perceived ease

of use (PEU) in influencing behavioral intention (BI) among students in planning to use Artificial

Intelligence as a decision making tool. This information is valuable when constructing user systems

because time spent on improving PEU might be more useful for the user and lead toward a stronger

behavioral intent of using the system if more time is spent of increasing the PU of the system.

The findings of this study may be questionable because that a nonprobability sample method has been

used when collecting the primary data of this study. However, because of the demographics provided

by the questionnaires filled in, the gender representation showed to be close to the original population.

Thus, indicating that the findings may be representative of the population to some extent.

6.2 Implications

It is important to be aware of that PU and PEU are people’s subjective appraisal of performance and

effort, and does not necessarily reflect on objective reality. In example, MIS studies has shown

discrepancies between perceived and actual performance (e.g., Sharda, Barr & McDonnell 1988).

However, Warshaw and Davis (1985) argues for that self-predictions, are among the most accurate

predictors available for an individual’s future behavior. Consequently, because there are few other

predictors of future behavioral available this measurement is perceived as acceptable in the model.

54

This study also focus on the perception of students behavioral intent not their actual behavioral intent

because the students are currently not using any decision making tools to make decisions in

organizations. However, it would be interesting to compare the result of this study with a study

conducted on a specific decision support system, letting students operating this system and therefore

have actual usefulness and actual perceived ease of use and comparing the result of that study to this

one.

Initially, it was hard to interpret and understand the complexity of the framework and how Davis

originally conducted his study back in 1985. Hence, many hours have been spent in trying to conduct

this study and test for reliability and validity as close to the original study as possible, within the given

time frame of this master thesis project. However, the sample size and construct of this study differ

from Davis that originally had 40 master students in his sample, and the sample size of this research

is 81 and the level of completed studies varies from high school degree, bachelor’s degree and

master’s degree. However, in terms of sampling, the researcher wanted the sample to resemble the

population in school as closely as possible, hence the variation in level of completed studies.

6.3 Suggestion for Future Research

A suggestion for future research would be to examine current decision makers’ in the region of

Gothenburg standpoint of using AI tools in order to make decisions in organizations. By performing

this, a specific decision support system tool can be examined and how the current decision makers

perceive this system may be found. These findings can be compared to this study conducted and

differences or similarities can be identified and measures can be taken upon the comparison of these

two studies. In what extent are students and current decision makers viewpoint of using DSS differ?

Why does it differ? What changes has to be done in the organizations in order to adapt for the changes

that will take place when the future decision makers enter the platform as decision makers of today's

organizations.

55

7. Bibliography

Agrawal, A., Gans, J.S. & Goldfarb, A. (2019). Artificial Intelligence: The Ambiguous Labor

Market Impact of Automating Prediction. (National Bureau of Economic Research, NBER Working

Paper Series 25619). Cambridge, Massachusetts: National Bureau of Economic Research.

http://www.nber.org/papers/w25619.

Ahlstrom, D. (2010). Innovation and growth: How business contributes to society. Academy of

Management Perspectives 24(3), pp. 11-24.

Amodei, D., Olah, C., Steinhardt, J., Christiano, P., Schulman, J. & Mané, D. (2016). Concrete

Problems in AI Safety. ArXiv, abs/1606.06565.

Bandura, A. (1982). Self-Efficacy Mechanism in Human Agency. American Psychologist 37(2), pp.

122-147.

Bell, E., Bryman, A. & Harley, B. (2019). Business Research Methods. 5th edition, Oxford: Oxford

University Press. ISBN: 978-0-19-880987-6.

Berlyne, D.E. (1966). Curiosity and exploration. Science 153(3731), pp. 25-33.

Boss, J. (2016). Staying Competitive Requires Adaptability. Online Source. [Available at:

https://www.forbes.com/sites/jeffboss/2016/04/26/staying-competitive-requires-

adaptability/#60549dcb7e6f] [Accessed: 2030-03-15].

Brief, A.P. & Aldag, R.J. (1977). The Intrinsic-Extrinsic Dichotomy: Toward Conceptual Clarity.

Academy of Management Review 2(3), pp. 496-500.

Brynjolfsson, E. & McElheran, K. (2016). The Rapid Adoption of Data-Driven Decision-Making.

American Economic Review 106(5), pp. 133-139.

Burgess, A. (2018). The Executive Guide to Artificial Intelligence: How to identify and implement

applications for AI in your organization. Cham: Springer International Publishing AG. ISBN: 978-

3-319-63820-1.

Camerer, C. (1998). Bounded Rationality in Individual Decision Making. Experimental Economics

1(2), pp. 163-183.

Campbell, M., Hoane, J.A. & Hsu, F-H. (2002). Deep Blue. Artificial Intelligence 134(1-2), pp. 57-

83.

Capgemini. (2017). Turning AI into concrete value: the successful implementers’ toolkot. Online

Source. [Available at: https://www.capgemini.com/gb-en/wp-content/uploads/sites/3/2017/09/dti-ai-

report_final1-1.pdf] [Accessed: 2020-04-29]

56

Carroll, J.M. & Thomas, J.C. (1988). Fun. SIGCHI Bulletin 19(3), pp. 21-24.

Cukier, K. (2019). Ready for robots? How to think about the future of AI. Foreign Affairs 98(4),

pp-192-198.

Davenport, T.H. (2014). Big Data at Work: Dispelling the Myths, Uncovering the Opportunities.

Boston: Harvard Business School Publishing Corporation. ISBN: 978-1-4221-6816-5.

Davis, F.D. (1986). A Technology Acceptance Model for Empirically Testing New End-User

Information Systems: Theory and Results. Doctoral dissertation, Sloan School of Management,

Massachusetts Institute of Technology.

Davis, F.D. (1989). Perceived Usefulness, Perceived Ease of Use, and User Acceptance of

Information Technology. MIS Quarterly 13(3) pp. 319-340.

Davis, F.D., Bagozzi, R. P. & Warshaw, P. R. (1992). Extrinsic and Intrinsic Motivation to Use

Computers in the Workplace. Journal of Applied Social Psychology 22(14) pp. 1111-1132.

Davis, F.D., Bagozzi, R. P. & Warshaw, P. R. (1989). User Acceptance of Computer Technology:

A Comparison of Two Theoretical Models. Management Science 35(8) pp. 982-1003.

Deci, E.L. (1975). Intrinsic Motivation. New York: Plenum Publishing Company Limited. ISBN:

0306344017

Djamasbi, S., Strong, D.M. & Dishaw, M. (2010). Affect and acceptance: Examining the effects of

positive mood on the technology acceptance model. Decision Support Systems 48(2) pp. 383-394.

Doshi, H.A.K., Balasingam, S. & Arumugam, D. (2020). Artificial intelligence as a paradoxical

digital disruptor in the accounting profession: An empirical study amongst accountants.

International Journal of Psychosocial Rehabilitation 24(2), pp. 873-885.

Field, A. (2018). Discovering Statistics Using IBM SPSS Statistics. 5th edition, London: Sage

Publications Ltd. ISBN: 978-1-5264-1951-4

Fishbein, M. & Ajzen, I. (1975). Belief, Attitude, Intention and Behavior: An Introduction to Theory

and Research. Reading, MA: Addison-Wesley. ISBN: 978-0201020892.

Gartner IT Glossary. (n.d.). Big data. Online Source. [Available at:

https://www.gartner.com/en/information-technology/glossary/big-data] [Accessed: 2020-03-15].

Gorry, G.A. & Scott-Morton, M.A. (1971). A Framework for Management Information Systems.

Sloan Management Review 13(1), pp. 55-70.

Gorry, G.A. & Scott-Morton, M.A. (1989) A Framework for Management Information Systems.

(SMR Classic Reprint). Sloan Management Review 30(3), pp. 49-61.

57

Green, T. (2017). A glossary of basic artificial intelligence terms and concepts. TNW News, 10

September. https://thenextweb.com/artificial-intelligence/2017/09/10/glossary-basic-artificial-

intelligence-terms-concepts/

Haenlein, M. & Kaplan, A. (2019). A Brief History of Artificial Intelligence: On the Past, Present,

and Future of Artificial Intelligence. California Management Review 61(4), pp. 5-14.

Johnson, E.J. & Payne, J.W. Effort and Accuracy in Choice. Management Science 31(4), pp. 395-

414.

Kacprzyk, J., Valencia-García, R., Paredes-Valverde, M.A., Salas-Zárate, M.D.P. & Alor-

Hernández, G. (2018). Exploring Intelligent Decision Support Systems: Current State and New

Trends. Cham: Springer International Publishing AG. ISBN: 978-3-319-74002-7.

Kahneman, D., Rosenfield, A.M., Gandhi, L. & Blaser, T. (2016). Noise: How to Overcome the

High, Hidden Cost of Inconsistent Decision Making. Harvard Business Review 94(10), pp. 38-46.

Kahneman, D. & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk.

Econometrica 47(2), pp. 263-292.

Lawler, E.E. & Porter, L.W. (1967). Antecedent attitudes of effective managerial performance.

Organizational Behavior and Human Performance 2(2), pp. 122-142.

Marwala, T. & Hurwitz, E. (2017). Artificial Intelligence and Economic Theory: Skynet in the

Market. Cham: Springer International Publishing AG. ISBN: 978-3-319-66104-9.

McKinsey. (2017). Smartening up with artificial intelligence. Online Source. [Available at:

https://www.mckinsey.com/industries/semiconductors/our-insights/smartening-up-with-artificial-

intelligence] [Accessed: 2020-04-29].

Oxford English Dictionary. (n.d.). Artificial Intelligence. Online Source. [Available at:

https://www.lexico.com/definition/artificial_intelligence] [Accessed: 2020-03-20].

Payne, J.W. (1982). Contingent Decision Behavior. Psychological Bulletin 92(2), pp. 382-402.

Pereira, J.L., Belo, O. & Ravesteijn, P. (2018). Special Issue on Big Data and Digital

Transformation. Journal of Grid Computing 16(4), pp. 531-533.

Rosenberg, M.J. (1956). Cognitive Structure and Attitudinal Affect. Abnormal and Social

Psychology, 53(3), pp. 367-372.

Rosenberg, N. (2004). Innovation and Economic Growth. Organisation for Economic Cooperation

and Development. Online Source. [Available at: https://www.oecd.org/cfe/tourism/34267902.pdf]

[Accessed: 2020-03-15].

58

Royal Society. (2017). Machine learning: the power and promise of computers that learn by

example. Royal Society. Online Source. [Available at:

https://royalsociety.org/~/media/policy/projects/machine-learning/publications/machine-learning-

report.pdf ] [Accessed: 2020-04-29]

Sharda, R., Barr, S.H. & McDonnell, J.C. (1988). Decision Support system Effectiveness: A

Review and Empirical Test. Management Science 34(2), pp. 139-159.

Sharda, R., Delen, D. & Turban, E. (2014). Business Intelligence and Analytics: Systems for

Decision Support. 10th edition, Harlow, Essex: Pearson Education. ISBN: 978-1-292-00920-9

Shrestha, Y.R., Ben-Menahem, S.M. & Von Krogh, G. (2019). Organizational Decision-Making

Structures in the Age of Artificial Intelligence. California Management Review 61(4), pp. 66-83.

Silver, D., Huang, A., Maddison, C.J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J.,

Antonoglou, I., Panneershelvam, V., Lanctot, M., Dieleman, S., Grewe, D., Nham, J.,

Kalchbrenner, N., Sutskever, I., Lillicrap, T., Leach, M., Kavukcuoglu K., Graepel, T. & Hassabis,

D. (2016). Mastering the Game of Go with Deep Neural Networks and Tree Search. Nature

529(7587), pp. 484-489.

Simon, H.A. (1955). A Behavioral Model of Rational Choice. The Quarterly Journal of Economics

69(1), pp. 99-118.

Simon, H.A. (1978). On how to decide what to do. The Bell Journal of Economics 9(2), pp. 494-

507.

Simon, H. A. (1960). The New Science of Management Decision. New York: Harper & Brothers

Publishers.

Swanson, D.L. (1996). Neoclassical Economic Theory, Executive Control, and Organizational

Outcomes. Human Relations 49(6), pp. 735-756.

Thaler, R.H. (2018). Nudge, not sludge. Science 361(6401), p. 431.

Turing, A. (1950). Computing Machinery and Intelligence. Mind 59(236), pp. 433-460.

Von Krogh, G. (2018). Artificial Intelligence in Organizations: New Opportunities for

Phenomenon-based Theorizing. Academy of Management Discoveries 4(4), pp. 404-409.

Vroom, V. (1964). Work and motivation. New York: John Wiley & Sons Inc. ISBN: 0471912050.

Warshaw, P.R. & Davis, F.D. (1985). Disentangling Behavioral Intention and Behavioral

Expectation. Journal of Experimental Social Psychology 21(3), pp. 213-228.

White, R.W. (1959). Motivation reconsidered: The concept of competence. Psychological Review

66(5), pp. 297-333.

59

Wright, J. & Ginsberg, D. (2012). Free to Err?: Behavioral Law and Economics and its Implications

for Liberty. Law & Liberty, February 16. https://www.lawliberty.org/liberty-forum/free-to-err-

behavioral-law-and-economics-and-its-implications-for-liberty/

60

8. Appendixes

A - Questionnaire Guide

61

62

63

64

65

66

67

68

B - Correlation Matrices

B.1 Perceived Ease of Use

B.2 Perceived Usefulness

69

B.3 Attitude Toward Using

B.4 Behavioral Intention

70

C- List of Exhibits

C.1 List of Figures

Figure 1.1 Overview of Research Process

Figure 2.1 Theory of Reasoned Action, TRA

Figure 2.2 Technology Acceptance Model, TAM

Figure 3.1 The Process of Deduction

Figure 3.2 The Process of Quantitative Research

Figure 4.1 Diagram of Technology Acceptance Model

C.2 List of Tables

Table 2.1 Concepts of Artificial Intelligence

Table 4.1 Cronbach Alpha Reliability of Measurement Scales

Table 4.2 Measurement Statistics of Variables

Table 4.3 Measurement Statistics of Factors Constructed from Variables

Table 4.4 TAM Regression Tests

Table 4.5 TAM Causal Regression Test

Table 4.6 Compilation of Adjusted R2 in Different Regression Models

71

C.3 List of Abbreviations

A - Attitude Toward Using

AI - Artificial Intelligence

ANN - Artificial Neural Network

BI - Behavioral Intention

DL - Deep Learning

DMT - Decision Making Theories

DSS - Decision Support System

KMO - Kaiser-Meyer- Olkin

OLS - Ordinary Least Square

MIS - Management Information System

ML - Machine Learning

PE - Perceived Enjoyment

PEU - Perceived Ease of Use

PU - Perceived Usefulness

SL - Supervised Learning

SN - Subjective Norm

SPSS - Statistical Package for Social Sciences

TAM - Technology Acceptance Model

TRA - Theory of Reasoned Action

UL - Unsupervised Learning